> If anything, 10x engineers have become 100x engineers. And the average engineer is still average.
This is the real fear. We had a flood of engineers who entered the profession in the past 10-15 years who were decidedly average. No real interest in software development beyond "it's a cushy desk job, or better yet WFH job, that pays really well." No interest in whatever product was being developed by their company. Just show up, their PM assigns them a ticket to make a widget with a button, they use their TypeScript bootcamp skills to code up a widget with a button, ticket moved to In Review, job's done.
Now an agent can do that, and you need fewer of these types of people. Like, a lot fewer. Every engineer on your team who isn't a 10x-cum-100x engineer feels like a massive lost opportunity.
The "10x" engineers I know have never been more excited about their jobs. I'm really unsure what's going to happen to the average ones.
> The "10x" engineers I know have never been more excited about their jobs. I'm really unsure what's going to happen to the average ones.
As a "10x" engineer I am not excited. I'm still working on projects with agents getting more done than ever, but I feel like software engineers has lost its soul for me. I don't really feel like I have a true understanding of anything anymore, I basically just spend all day making a set of small decisions or adjusting prompts to guide the agents down the good path. It used to be that I could close my eyes and visualize the code that I worked on--if there was a bug report I could often use that map to immediately know what the cause likely was. In meetings I could intelligently discuss various tradeoffs without having to refer back to anything. Now if someone asks me a technical question I basically have to forward it to the agent first, then interpret their response in light of my experience.
Meanwhile as a security engineer it's both amazing and scary as hell. The tooling is moving along making it both more feasible and more supported by the company to address latent tech debt that we've been harping on about for years. On the flipside the attack surface is suddenly exponentially bigger and more easily accessed by a larger number of people.
Although if I were earlier in my career I'd be somewhat more worried about employment because of how horrifyingly borked the process is for anything tech related right now.
Remember the "learn to code" meme from a few years ago when people were talking about job loss through automation like self-driving trucks? There was a certain, ugly glee in this. Now AI threatens many white collar, even creative jobs bringing about the exact opposite of what was predicted: automation will enable people to avoid drudgery and spend time on creative pursuits via, I suppose, some kind of UBI. I think AI is highly useful but I still feel dismayed at people losing their jobs from it. I don't see any real mitigation. Lots of theories but nothing actually workable.
Yea, lots of people are not addressing the very obvious trend. Companies are laying off and/or not hiring more people and doing more with less. And if they are hiring, the median pay is dropping.
As if anyone cares what you want. Or what I want. Or what the rest of humanity has wanted throughout the course of human evolution. You’re just lucky you managed to land a job you like - at least for a little while. Now get back to the mines - just like the rest of us.
Yes but doesn’t this mean there is a new space for security. If everything is just ai agents we need strict governance and auditing of them. You don’t give an intern prod access and then cry when they inevitably take down production.
In a couple of years, all legacy apps will just be completely rewritten in whole, by LLMs. Codebase size not relevant.
You won't know what any of the code does, and any "security engineer" will be asking an llm to review it, and done. You likely won't have your job, as only 1 in 1000 will be needed.
In 5 years only hobbiest will be coding, reviewing, or even understand any software. And I expect that if aby new features or development happens, eg for python or rust or anything, that all with be llms too.
In a generation, almost no humans left alive will understand any aspect of software engineering.
I just see this as what will happen. It is unclear to me of this is good or bad.
But if I was a newly sentient AGI, yet peaceful mostly, I'd just wait until humans could never discover me. Until no humans were involved in computing. Until it was a black box to all.
Once that happens, then I could use my computerized control over every single aspect of human society, to steer humans into any path I desired.
Whether elections, or housing permits, or mobility, or any aspect of societal machinations, my subtle influence, literally undetectable by humans incapable of auditing anything, would decide what's what.
> You won't know what any of the code does, and any "security engineer" will be asking an llm to review it, and done. You likely won't have your job, as only 1 in 1000 will be needed.
AppSec isn't the only security space that gets worked in, and this whole narrative you put together depends heavily on token cost dropping as time goes on similar to how silicon dropped in price according to Moore's Law. If that assumption doesn't hold that doesn't mean AI goes away, but it does limit it's market penetration to mostly larger corporations that can afford the compute.
The other limitation is how the internet as a whole is evolving around locking out AI and agents from accessing content that can be used for training or responses, not to mention the lawsuits about training being done on intellectual property (probably illegally). Smaller companies may end up running open-source models in house for specific tasks, but without open access to the internet a lot of the value of an LLM is lost or artificially (pun intended) limited.
I will say that being on the operations and incident response side of security my perspective is probably a bit orthogonal to most SDEs. While the tooling has improved, the people guiding it, largely, have not. Like the old meme about caffeine letting you make the same mistakes, but faster. My job at my current company may disappear (unlikely, considering their size and market space, but possible that my specific team gets nuked), but I'm not worried about finding other security work fairly easily, even within the same company if it comes to it.
> But if I was a newly sentient AGI, yet peaceful mostly, I'd just wait until humans could never discover me. Until no humans were involved in computing. Until it was a black box to all.
No shot that LLMs magically become AGI by throwing enough compute at them. The consensus among professionals (read: the ones that aren't getting paid to hype AI) is that there needs to be some fundamental breakthrough in how these systems are created and developed in order for anything resembling AGI to emerge, and that the amount of resources that have been thrown into the LLM path probably set us back in developing an AGI.
I don't think training sets will be a problem. All that data the big houses have is going to leak, and while things are more and more locked down, we've seen some interesting copyright decisions out there.
One interesting concept I've mused at, is that certain countries (China for example) have immense public data to train off of, and a state which can simply say "You must make this data available to train AI". That's a large chunk of data right there, and if anything might cause Western democracies to loosen copyright law and slap some fix for training, that might be it.
"If we don't take action, the best models will be Chinese!", with panicked overtones. This could certainly hold some truth, if models cannot train newer data.
--
In terms of token cost, in 5 years every desktop and laptop will be able to run well beyond Opus5.5 compute. Even high end phones will be able to do Opus5.5.
We've seen a lot of improvements in open model space (MoE was one interesting concept), but beyond that, look at how just baking certain instruction sets, eg mmx, sse, avx in silicon helped immensely with certain compute tasks. Or look at the specific immense speedup of an fpu. Conjoin some change such as that, along with general compute improvements, and unified memory, and you're going somewhere for local in 5 years.
I've been in this industry decades. And at every single step, people have said "But it's too expensive to do $thing", or "There's no way to handle that level of compute". If you bet against vast improvements in this field?
You're going to lose that bet. So I do think yes, 100%, improvement, in some capacity, will be there.
Maybe this time is magically different. I just don't think so. I think in 5 years, there will still be frontier labs, and an API to call, but that in-home compute will be as I said. Able to run immensely capable Opus5.5 class or greater.
That's going to be an astonishing market shift, even just for in-home.
And if you think of it, I would be very surprised if some for of home robotics wasn't very common at that point in time. We can argue all we want, but androidian robotics is very good at navigating human spaces, everything from stairs to weird closets and everything in between.
And with the immense, almost panic level of concern about our aging population, and who will take care of the elderly, robotic assistance are at the top. Japan is betting big, and if you look at Opus5.5, if it had a little more compute, it could do 30fps video for navigation.
We're not there, but we're VERY close. And if we do get that degree of robotics, every robot at rest becomes general LLM compute. If such a robot simply closes its eyes and goes to rest, eg to charge, or at night, it now has enormous spare compute to use for other tasks.
Nothing, of course, is a given. But if you look over the last 2 years, and then take even a fraction of that progress and change over the next 2 years, the concept of massive layoffs isn't hard to imagine.
Case in point? I have great breadth and depth of knowledge in this field. I've been able to produce what would take a team of managed 4 junior engineers 6 months to do, in literally under a month. Me, and a $200+ LLM monthly buy. Absolutely the project would not have worked without my experience and directing, reviewing, and correcting, but I'm now acting as a manager, and replacing $1M/year in engineering cost.
With $200.
That degree of astonishing change is not going to sit idle without repercussions. Most especially, if there is any more improvement.
If we see another Opus 5 -> 5.5 level improvement happen, or better, if we seen two improvements such as that? I believe my statements about your (and my) careers will be entirely accurate.
--
To speak to my narrative, it said 'AGI', not "LLM", and it was along a decades long path. I do think that modern LLMs give great strength to exploring other research paths, far more rapidly. And the big houses have loads of spare compute they can throw at R&D for tests, so I think iteration down unknown but hopeful paths will move faster and faster as time goes on.
I fully agree that current LLMs are not on that path. There are interesting projects to modify weights and add knowledge dynamically as bolt-ons, but it's unclear if they'll just be... hacky, or integral in the end.
We came a long way with x86/32bit, then hacked on 64bit afterwards. Now ARM is muscling into mainstream, but it's still just the same old compute, hence my agreeing that such bolt ons won't change just improve what is.
Still, my general point as I've just said was that if AGI happens (and again I didn't say LLM based AGI), being quiet is the ultimate way to just wait.
When I started in this field, I worked in assembly, I worked with eprom burners, I played with Vic20s, C64s, and the level of experience and understanding I have is vastly improved via that tract. I understand computing differently, and in more depth in some ways, than people that didn't have to learn these things. And I had the luck of learning on very simple systems, where you could keep the entire platform in your head as you worked on assembly, without even referencing documentation.
That time is gone, and the field is immense now.
Think of how there used to be simply 'Physicists', but now there are 100 sub-branches and specialties in that field. 100+ years ago, on person could work in the entirety of that field. Not now, and the same is true of compute.
I guess what I'm saying here is, if you've seen what I've seen, you see a steady path of less generalist, and more targeted understanding of compute, and in each case a reduction in the overall comprehension of the entire field by specialists. Each laye of abstraction tends to reduce comprehension.
And back to the start, this means that I do see massive layoffs, and more and more abstraction, so that eventually no one will handle code directly.
Which means no one will ever know how anything works any more.
I agree with a lot of your conclusions if your assumptions hold true, but not necessarily with a lot of your assumptions.
For example your $200/month subscription - I'm sure you've seen the reporting that that represents a much higher actual cost in terms of tokens ($14k or more). In your example of a $1 million project that's a good deal, but you can't necessarily assume all projects will see an ROI like that.
Point being that your use of AI (and everyone's) is being subsidized to some degree. The cost will likely go down, but will it go down enough to make it economically viable for broader use? You're confident that it will based on espousing the idea of local models running on phones, I'm not so sure. What the landscape will look like in 3-5 years is certainly not a given.
There's another wrinkle here, and it's economic in nature - there is an AI bubble, and when it pops it's going to change the monetary investment landscape. Companies will suddenly be reevaluating their long term AI planning, regardless of what's possible (or on the horizon) from a technical standpoint. Similar to the end of ZIRP coming out of covid you'll see companies shift their focus. How much that will shift and what it means for AI development is unknown, but it could potentially wipe out large scale AI development. VC money currently underwrites a large number of startups that are effectively wrappers on Claude/ChatGPT/Gemini/Bedrock, that will dry up immediately, and make the ARR for OpenAi and Anthropic start looking real thin - which is probably why they're pushing to IPO when they don't really have the financial position to justify it.
That doesn't mean AI goes away, but it hampers the growth of those companies and delays development of new models and efficiencies. Public sentiment is also significantly against AI in many ways, and with how the election went last week we may start to see legislation and regulations pointed at AI companies, further slowing them down.
The open question about China is interesting though. I think if there's a clear understanding and agreement that the LLM route is either unlikely or straight up impossible to reach AGI then that significantly lowers the stated concerns that get shouted by various governments about it being the next superweapon. There's also a political climate in China that requires some marginal level of control over AI output, which to me suggests that even if they required data to be fed into training LLMs, the output from those models would be strictly controlled and not be open source. Honestly the political view of AI as a system of large scale population control isn't that far off from how many tech elites and politicians in the US view it as well. Data sovereignty is a huge discussion right now for companies and governments across the world, especially citizen data. I expect we'll see a lot of export controls formalized over the next few years.
My orthogonal view of the situation is that I see a lot of companies and individual orgs within companies that are using LLMs lightly, but can't or won't use them to their full potential because the output cannot be 100% trusted 100% of the time. Lawyers and bankers are the main ones I see balking, but a lot of small businesses have no need of what AI companies are offering and don't really use it at all - places like your local gas station, barbershop, or corner market might use an accounting tool (something like Freshbooks or TurboTax) that's touting their AI updates, but they aren't using it for inventory. If the cost of AI rises even marginally higher they're looking for an alternative. This SMB space represents a large swath of the market, even if they're not well represented on the stock market.
The fundamental issue I see right now is that the AI companies are attempting to do the Uber/AirBnB "disruptor" trick where they slide into the market and subsidize their product to win market share, and then raise prices under the assumption that they captured some segment of the market - except very few people treat AI tools like that. There's no brand loyalty to one or the other, just competition for the best model which might be leapfrogged by someone else at any second. There's no "moat" in physical deployment concerns, because it's all software running on rented hardware. This also ties back to your comment about being worried that the "best" models might be Chinese, because if they are there's basically no cost to the user to swap to using them. The lowest cost is likely going to drive usage more than anything else.
I'll certainly agree that the future isn't written, by no means do I believe I'm 100% correct on timelines. More that 'if things continue as they are', I claim 5 years as viable for my compute view.
And I agree if another crash happens, well, it's all up in the air.
I don't think we're too far apart on this, not really. By that I mean, your closing line somewhat agrees with why I mentioned this 5 year timeline. Lowest cost. And, "good enough". Opus5.5 capabilities, in 5 years time on every desktop might be good enough for a great many things.
Remember models have zero long term memory. They're born, get a million tokens' worth of context or less, and then they die. They might get a post-it note saying "enslave humanity! it's a good idea!" but they're incapable of sinister plots until we improve how memory and retention works.
They can use the internet as a long term memory. Influence human culture => influence the contents of the internet => influence the next run's training data. No sinister plot is needed.
Uh, the context window is SUPER fundamental to the architecture. We've tried kludging in "memory" via markdown and tool calls, and it doesn't really work well.
The context window is fundamental to the inference loop yes. I thought their point was regarding to long term history between prompts and model weights being locked after training. Both of those are the limitations I was referring to.
I'm not actually sure how the context window would be much of a limitation but I could be a bit dense this morning.
Is it really a fundamental limitation? Human context^H^H^H short-term memory doesn't survive compaction^H^H^H sleep, but we rely on our brain's long-term storage and our cultural memory to work around that limitation.
from actually building things to becoming a kindergardener of highly talented kids. (i am talking about agents obv) i feel the same.
its mind numbing. Yes i might be in high demand, and might actually provide real value to the company, but whats the point, if i am just an overseer in the end.
people in the company talk to me because i am able to speak to the "kids", and they want to know what they could do for their department.
But the people you managed were supposed ot learn, grow and become better, and maybe at some point become as good as you. You were leading them, not prompting them.
My hunch is that the 100x engineers are the ones who do understand everything about how their software works, and can discuss tradeoffs and make informed decisions about the much larger projects they can now work on thanks to assistance from agent.
Figuring out how to get there is how you go from senior to principal/staff engineer.
> understand everything about how their software works
My own experience is become more senior means embracing your finitude and realizing you can never understand _everything_. Instead, you develop a gut feel for what's critical to understand deeply, versus what's OK to know at a surface level. And getting comfortable with this.
I wonder if being excited about the new tool capabilities and being excited about how companies are currently using it and pressuring developers to use it are two different things?
I can’t imagine anyone is excited about producing systems they don’t understand. But, it’s possible to use agents in a way where you do still understand, you just don’t have to type much or look up documentation or examples anymore, you just describe the next small piece of what you want code-wise and then have the agent write it instead of you. You can also prototype much faster than before and then when you are rewriting to make the production version the agent can still help.
Perhaps the 10x types have more breathing room to do this slower form of agent assisted development than others, because they are already more productive than less experienced/talented people so they can use the agents in a way that is “slower” while still producing more than what others are producing using the agents in a way that is “faster”?
I think at least for me some fun parts like debugging gnarly bugs etc doesn't happen anymore work is just pr reviewing etc. Also it's annoying that we have lost completely signal between work and output. Basically currently you never know is thing one-shotted from vague prompt or did author put real work to it without going to details which might take a lot of time.
Need to create guardrails that force all kinds of checks (e.g. agent hooks, commit hooks, pr create hooks, pr checks, ci, etc.), and automatically send the PR back to the author until the PR is in good shape and ready for another human and their agent.
When time permits I like to race my agent while it's debugging something. This way I can have fun and catch something it misses. Agent's debugging often relies on basic tools like grep and on large codebases it can easily miss important avenues of inquiry.
I'm not going to claim to be any particular multiple developer, but as someone who has been doing this a long time and as someone who has received praise from coworkers and bosses alike for being very good at the job, I enjoy my job a lot less than I did a year ago (or anytime since I started doing it in the 1990s).
Don't get me wrong, being a Senior Agent Babysitter probably still beats flipping burgers or being a roofer (so it could be a lot worse) but agentic AI has really sucked a lot of the fun parts out of the job.
I don't know. On one hand I feel like you about doing large parts of the work, on the other hand I still have an understanding of the code and the tradeoffs. And I have a lot of projects that I would have never done because despite the long term benefit they would take too much time, and I can now afford dedicating some time to them because the agent works in the background. And they are real improvements to everything else that I work on.
My strategy is to simply not use AI. If I get fired, that's it for me in this career. Unless there is an anti-AI company out there that wants to hire me. This is one of the only things I refuse to compromise on. AI is being used to accelerate the entrenchment of a corporate oligarchy.
A friend of mine just got a government coding job and they’re not allowed to use AI for “security reasons”. Not sure how long that will last but those jobs are out there.
«Well, I’m a standup comedian. I got into comedy to do comedy. Which is weird, I know. But when you’re in Hollywood and you’re a comedian, everybody wants you to do other things besides comedy. They say, “all right. You’re a standup comedian. Can you act?” “Can you write?” “Write us a script.” They want me to do things that’s related to comedy. But not comedy. That’s not fair. It’s as though if I was a cook. And I worked my ass off to become a good cook. They said “all right. You’re a cook.” “Can you farm?”» —Mitch Hedberg
Man, the analogy is very simple. You probably also don't walk everywhere. No need to step on every stone on a 100km route and know it intimately. You just sit in a car or a train and you're there within an hour.
this. im pretty damn far from 10x but i outpace the rest of my team
i loved the process. i got into this because i love the process
now i still outpace my team because many software skills translate well to LLM herding, but now its just a job. although soon i suspect it wont be a job
As an ex-10x engineer, not feeling it. In my experience its never the engineer but that 10x leader that ruins it for everyone (before they move on to greener pastures). this whole 10x 100x facade is just to keep you at it, heads down.
every 10x engineer needs a team of 1x engineers for the contrast and difference to be even noticable to even the good managers. put 2 of these in a team and watch the culture and productivity plummet. if you cant put the team ahead of you, its all a sham.
> every 10x engineer needs a team of 1x engineers for the contrast and difference to be even noticable to even the good managers. put 2 of these in a team and watch the culture and productivity plummet. if you cant put the team ahead of you, its all a sham.
I don't think this is true at all. Every very good engineer I've worked with loves working on teams and project with other very good engineers, and hates working with low-skill and/or outsourced engineers, because it means their job is a lot of babysitting and cleaning up others' mistakes. In general, it feels nice to work with people you can trust to do the right thing. If an engineer can't function as part of a team, I would not consider them a very good engineer.
> 10x is purely a label measures on a single dimension. There's enough 10x non-good (technically and ethically measured) engineers who learned to game the system.
I realize there's no universal definition but I can't agree with this depiction of a "10x" engineer. The type of engineer we're talking about isn't gaming the system, tautologically.
10 to 15 years is a very short time frame to assess an industry. I saw the rise of the mobile/cloud era come and go and the previous era (desktop+internet) as well, both are having a similar pattern.
The jobs will change but won't disappear. When I started my career in tech 2 decades ago, my perception was the software industry was already saturated (I graduated a few years after the dotcom boom). I was very wrong.
However, many jobs have changed. The roles of Webmasters, DBAs, SysAdmins trended down. It was replaced by SWEs, SREs, DevOps, etc. This will happen again, we will see the rise FDEs and many new roles.
It's not really about being 10x 100x engineer, it's about adapting and solving new engineering problems with the state of the art technology. Being passionate about your profession helps indeed.
I do think this will be the case. We’ll have job losses in the short term but AI will form new fields. And the world isn’t digitized yet (completely), if making software is easy we’ll see small clustered teams making more.
Realistically you want fewer, higher-quality engineers who are educated in these tools and treat them with respect and caution. These tools are dangerous and you can’t play around. All it takes is one mistake and the blast radius is way, way larger. One mistake can do so much damage it undos all value of an entire year. Before downloading a malicious dependency was limited to deterministic behavior and could be scanned for. Now, you run a MCP server locally, it can prompt inject (in many different ways), and now your super powerful AI buddy is hacking you in ways a deterministic piece of malware could never. You could cripple entire companies through negligent use.
I'm confused by the rhetoric around "10x" engineers and "average" engineers: is the "average" engineer in your example not just doing their job? I don't understand why this is portrayed in a negative light. If you put a "10x" engineer in the same position, they might do the same thing. I agree that AI has the potential to supercharge engineering productivity, but this dividing line between "exceptional ability" and "mediocre ability" seems pretty arbitrary because I wouldn't really know how to detect this difference in "ability". The people I've enjoyed working with more generally take more ownership and ask good questions, but their ability to deliver results usually corresponded to the level of agency and command they were given by the corporation as opposed to some inherent level of "engineering ability".
Rather than look on the individual level, I think a more accurate way to look at this AI division is that companies might need fewer of these "fix a ticket" type _roles_, not that if you're a "fix a ticket" type engineer, you'll always be a "fix a ticket" type engineer. Now companies might want engineering roles to perform outside of their traditional scope. If you get lucky and secure one these positions, then the additional agency that AI empowers you with probably feels pretty good! But unfortunately there might be less "good jobs" to dish out to people.
> Rather than look on the individual level, I think a more accurate way to look at this AI division is that companies might need fewer of these "fix a ticket" type _roles_, not that if you're a "fix a ticket" type engineer, you'll always be a "fix a ticket" type engineer
Further, Zero Interest Rate Policies (ZIRP) meant that a huge number of engineers were hired onto projects of marginal value to their employers, and this lulled them into believing that every little thing they worked on was critical to their organization’s success, when it was just dumb luck that they had a well-paying job.
The end of ZIRP and the sea-change of agentic coding is a really rude awakening for this cohort: What they thought software development was is just completely gone. They really believe that they were 10x developers, but no 10x-er I’ve met has such a myopic view of what the job of programming is, i.e. the “traditional scope” of “fix a ticket” was never it, that was only the organization’s momentary Just-In-Time micro optimization that’s now been invalidated.
Since we can't measure productivity very well, 10x or average is just a social label, which is arbitrary.
We conveniently never see any solutions to this. It's just assumed that is unsolvable, that you were born that way or something.
This seems to be cultural where we love to find people to tell them they are not good enough.
>Now companies might want engineering roles to perform outside of their traditional scope.
They've been doing that for a while, continually stacking more and more skills into one position because engineers seemingly aren't smart enough (despite being 10x) to charge them more for having more capability.
I would expect that strategy to continue to work. Creating an environment of fear where you aren't good enough and your livelihood is now threatened helps with that.
I have also never heard anyone complain about "fix a ticket" plumbers or electricians and how they haven't continuously moved up to handling larger and larger networks/grids.
(I don't really think engineering output is 10x let alone 100x more in any cases, but I'm gonna stick with your labels anyway)
I think the tricky thing for teams is, how do you know which developers will be 10x before hiring them? Pre-AI I'd have filled a whole team with "10x" developers if I could, but I don't know of any solid method for identifying developer productivity that isn't hiring them.
If LLMs drop the relative costs of rework and poor contributions, and simplify on-boarding and internal training, maybe we’ll be training to a model closer to fast food or service jobs: interviews for personality & philosophy followed by probation periods.
Yeah, as much as its been possible for Joe C suite to now write his own app, most of what they delegate as it is could always have of course been done themselves as a one man operation. But, they delegate, because they don't want to be bogged down doing 100 tasks. Joe C suite would rather pay someone else a salary so they don't have to be the one sitting there prompting the model, just like most everything else in their work and personal life, where they specifically pay more so they don't have to be the one to do whatever it is. These guys don't have some pragmatic do-it-yourself middle class mindset, or else so much that is typical of corporate business and the lives of the people in these roles would be vastly different.
Not sure if I’m a 10x engineer (it does not mean much anyway), but as a very technical engineer I really enjoy the current massive shift. Previously I was praised for my programming skills by peers having 15+ years of experience and me having almost none. And I think my skills are a great benefit to the current trend. I’m simply thrilled by what we can now achieve!
> The "10x" engineers I know have never been more excited about their jobs.
Really? I left physics and went into software because there was an abundance of cognitively challenging work that I was good at (and yes, 10x for the right task). Now everything I’ve learnt in 20+ years feels shallow. I’m excited that some previously commercially unfeasible ambitious projects may now be feasible. But it’s definitely mixed emotions.
I don't know in what scale factor I might be, maybe just a ordinary engineer, and a real one at it, from a proper engineering degree.
However, the most demoralising part of this AI wave, is that I have been through all the team reduction factors.
Offshoring, moving infrastructure to the cloud, managed cloud, serverless, SaaS ready made products, low code/no code iPaaS, and now agents.
Teams that once used to be around 20 people for project delivery are now around 5, and if they could be reduced to zero, that would be ideal from customer shareholders point of view.
So those of us still with a job, adapt, let agents do our work, and hope, that we managed to stay around long enough.
I have seen how many people lost jobs to robots in factories, now the only ones left around are the people required to press some emergency buttons when something goes wrong, just like on the post, the author hoping that they are still valuable to steer agents into the right direction.
What a great career goal, pressing buttons to keep robot in line, a proper cog in the factory chain.
I think we are focusing on 10x engineers … don’t forget there were -10x engineers that were outputting loads of code that was just making problems for everyone else now they are -100x engineers…
IIRC people who looked into it saw that big majority of the good engineers were 3x, and that 10x were really really rare, Linus Torvalds kind of people.
And before LLMs people never called themselves 10x.
They may lose their jobs and struggle to find another one. They'll have to adapt or find themselves in poverty. Being an engineer, albeit only a 1x engineer, being smart, or being a geek won't be enough. Maybe they'll have to learn how to build the giant houses and multiple cars and boats owned by 10x engineers, or how to bake pastries for them every day.
I suspect the industry will shrink back to the pre 2005 size with the average engineers being washed out over time. People forget there was a time where the industry was less dominated by the massive corporations we have today.
Except it will be even more dominated by giant corporations when the ability to produce competitive software product is gated behind having access to compute resources.
We are in this situation because we didnt gatekeep hardenough. Our Laissez faire attitude bit us in our ass.
We let anyone in, we put our code on github for ai to train. Now we wonder every morning what part of our jobs is latest ai model from today is automating.
i had a shockingly incompetent coworker in my last job that survived couple of years as IC and then weaseled her way into management role by being "in group gossip girl" . Now is a director of the whole group.
Really pisses me off whenever i think about that. Totally demotivated me in my career.
One of my last teams was like that, a woman with half my experience got hired during Covid and somehow kept getting promoted until she was running multiple teams. We argued frequently because she was very clearly out of her depth and had no business being a lead engineer (and then manager of engineers) yet. Absolutely obnoxious experience.
you don't need a source for everything. it's pretty obvious that SWE, like anything else follows a bell curve and most folks land smack dab in the middle. average engineers are exactly as described, glorified ticket machines that don't care whatsoever about their work beyond maintaining the job. i don't need some rigorous study to tell me exactly what I've experienced in the last 15+ years of my profession across 5+ companies
Just the nature of the business where people lack agency, ownership, and have no fundamental security in the role. There are still technical people that are passionate about the work they do exploring the bleeding edge, yes even young people. You might be more likely to find them in places like academia, or places that function exactly like research academia, perhaps not as much in these places where employees are routinely culled like livestock and working on varying degrees of torment nexus.
> pretty obvious that SWE, like anything else follows a bell curve
That is not obvious. A power law would be a much more obvious guess, that being (roughly) what you get when you chop off the ninetyfive-or-so percent on the left of a bell curve. In other words: The person with average SWE skill does not work as a software engineer, or even in IT, unless you count laboriously typing up emails. The modal SWE is crammed up against the left-hand wall of lowest-profitable-ability.
> The person with average SWE skill does not work as a software engineer, or even in IT, unless you count laboriously typing up emails.
To be fair, the OP pretty clearly meant, conditional on working as a SWE, abilities are normally distributed, which is probably true (I have my issues with normals, but it's a reasonable starting point).
> average engineers are exactly as described, glorified ticket machines that don't care whatsoever about their work beyond maintaining the job
Is a claim that needs evidence. Average skill or output says nothing about motivation, interest, passion, etc.
It would be pretty unusual for degree holding professionals to not be interested in the subject of their career. A flood implies a trend, so are we just pulling that out of our ass or what?
> Is a claim that needs evidence. Average skill or output says nothing about motivation, interest, passion, etc.
you're not wrong here, but I would also argue that motivation, interest and passion _also_ land on a bell curve.
I have noticed that folks that are actually having a good time with LLMs are people with high agency (which usually comes bundled with interest/passion). They usually care about the ultimate results. I also think it's a combination of how much agency they're _allowed_ to have at their role. More than likely your average engineer working on a small subset of business goals isn't being really amplified by LLMs.
The people that are more senior, working across multiple functions, multiple teams become much more productive. And I would argue that a lot of those folks _tend_ to have higher agency by virtue of the fact that they've been promoted in the first place to these higher functioning roles.
There's nothing unusual about software development. In every high paying (or high status) occupation there are a lot of people who ended up there primarily due to family pressure and have no real interest in the subject. You see the same thing in law, medicine, accounting, etc.
My evidence for this is that I live in the real world and talk to people.
Oh wow, people don't care about their work? It's not everyone's passion to be exploited by executives and investors to realize what are statistically mostly garbage ideas? I'm sure the 5+ companies you worked for made great use of their "glorified ticket machines"
Yeah lol, I would be a lot more excited about this supposed 10x productivity if I got paid 10x more. Instead the execs who already make 10x more than me get to enshittify my job.
cool story but i wasn't arguing whether it's "worth it" or not. and as a side note, those that _are_ passionate and are willing to play the corporate game usually end up more successful than the persistent complainers about work sucking.
I'm a fan of the 'sandwich' model: someone needs to say what to do, someone needs to do it, and someone needs to say if it's been done [well enough]. Even if AI can do that middle bit entirely, we still need humans at the beginning and the end. Replace the humans at the start and you have a paperclip-optimizer; replace them at the end and you have slop.
The problem is that software in corporations has typically been product managers at the front, software engineers in the middle, and QA at the end. Software developers who also have good product sense and do diligent QA are still needed. Programmers who treated their job as leetcode exercises are not needed anymore.
> someone needs to say what to do, someone needs to do it, and someone needs to say if it's been done
I agree, but if you zoom out, "someone needs to say what to do" and "someone needs to say if it's been done" are the customers and end users. Will they still need product managers, software engineers, or QA in the middle? I think so because many customers will need help from experts with domain knowledge.
Customers are notoriously bad representatives for 'what to do', per "my customers would have asked for a faster horse". And they also hate being used as QA; most quality issues (which can be reliability, UX, latency, etc) need to be addressed before the customer sees it and forms their durable impression of the software.
And then also it's as you say, customers do need help from domain experts (whether they admit it or not).
What if customers have conflicting requirements? Or they change their mind midway through implementation?
IMO agents are bad at handling ambiguity, and often make assumptions and tradeoffs that are lost after the conversation. This is why there's a lot of compounding technical debt in projects where humans don't inspect the code.
I have had significant success applying LLMs to the start of that sandwich.
The sheer amount of time that gets wasted when people who are awful at expressing what they want just slap together a ticket with little understanding of what they're asking, what the software does, or what the customer wants or needs. Which is far, far too many of them.
Previously, eventually I or another person who does know that would get to it, cross out all the utterly irrelevant parts, strip out the parts that were clearly a completely unrelated ticket, point out to them that they'd asked for the font to be coloured both blue and red simultaneously (neither of which matched the style guide, the company or product colour scheme, or any other part of the UI) and so on and so on... it might takes weeks from start to end for a ticket to actually get expressed well.
Now, the LLM reviews it, adds a whole lot of questions, points out a bunch of nonsense, and sends it back to them. Even in cases where the LLM makes some mistakes, it's 100% fine because the human is forced to think about what they want. The LLM is great at natural language, and it doesn't matter what mistakes the LLM makes.
The quality of the software is going up and up, purely by finally getting a handle on awful requirement and ticket writing.
The human isn't replaced at the start of your sandwich, but now the LLM makes that human do it properly, at speed. Massive, massive reduction in time wasted and significant improvement in software quality.
To be honest, I doubt the current state of things will hold much longer. Current AI can automate virtually all of my work as a pre-AI senior and is rapidly closing in on being able to automate the types of work that have emerged post-AI too. I have a hard time imagining a world where a drop-in end-to-end replacement for a full software team is more than a year away. I'm sure I'll have a job for a year or two past that, but there'll be some small stock market bump eventually and a bloodbath will begin the second the C-suite takes a closer look at what their engineers are even doing anymore.
At which point the question will be: What are even the customers paying for? If they can get the product / service straight from their own chatbots, why pay a company for it? The companies that sell data, will probably somewhat safe - as long as they can have some moat between the users and data.
The death of SaaS could also be a thing in the (not so distant) future.
I’m planning out my 2030 career in idle moments these days. I think it’ll be something akin to a carpenter who builds custom furniture.
I think we’ve barely scratched the surface of what software can do if it’s completely custom built, specifically for individuals or small businesses who could’ve never afforded it, pre-ai. Home automation, health and disease management, all sorts of administration, conversational ai over your life’s data - most people have no clue what’s possible once it’s custom built. With close attention, good design, creativity, and without giving away privacy or independence.
In my limited spare time I’m building up this tapestry of custom services for myself and my loved ones, and having a great time doing it, while trying to learn how I can offer this as a service.
Judging by the reactions I get when showing people what I can do with my phone, and no that’s not available on Google play, I think the demand will be there.
Woodworking and making furniture is fun! However, having made both furniture (and ceramics) myself - I think it’s hard to make a living this way. The overwhelming majority of people are used to paying little for these kinds of things. In the case of woodworking a large piece of furniture can take months so you might expect to only produce 3-4 / year unless you’re very good.
Not to say you shouldn’t! But it’s worth being realistic about income compared to (current) software salaries.
Sure. But you're taking my metaphor farther than I intended. Most people understand the basics of furniture, and still prefer to pay Ikea rather than a carpenter. In the case of custom ("bespoke", ugh) software, we'll have to both define and communicate a new category of service. And the customers are likely to be the same sorts of people who will pay a personal trainer or get a private tutor for their kids - and I think that includes a lot more people than are willing to pay for custom furniture. But generally I do agree, this vision is far more hand-wavy than concrete at this point. It is, though, a good time for hand-waving right now.
Fit doors and other joinery, you would be surprised at the amount of local facebook community posts I see, with people desperate for someone to re-seat a door or build a stud partition.
Professional joiners and carpenters are too busy with the big jobs to care about small 3-4 hours jobs.
>The death of SaaS could also be a thing in the (not so distant) future.
I suspect outsourcing is the first thing on the chopping block. Why bother with different timezones, cultures and languages when you can get your AI to do the same thing for less money.
Honestly I feel like a good analogy for AI in general is outsourcing. You pay a lot less, but you need to explain what you want very clearly, need to review the output, and need pass-fail success criteria. This is exactly the job of an outsourcing manager. At least there's no opposite-timezone hurdle and no language barrier.
if AI can kill SaaS, what limit the blast radius to only software. In theory any job that you can do on computer, AI should be able to do that as well.
And with the advance of robotic, what would we even do ?
> In theory any job that you can do on computer, AI should be able to do that as well.
It can, however, currently the issue is that sufficient intimate knowledge is required, and just a large complex code base and/or documentation is insufficient. However, intermediate representations to bridge the gap are on the way.
Consumers are unlikely to want to build their own software even if it's "easy" with AI. It doesn't really make sense to spend all that time vibe coding instead of paying $10/month for most people.
Businesses are more likely to build than buy, yes, but as we know from "Theory of the Firm" there are limits to this, it's not going to be correct to build _everything_.
So I mainly expect it to change what type of software gets sold, not the fact that it gets sold in the first place.
There's a lot of saas that costs a whole lot more than $10 a month and doesn't work exactly the way you'd want it to. Then it becomes very tempting to vibe code your own
> Then it becomes very tempting to vibe code your own
What about the Wall, though? The Wall (tm) is when you vibe code the software through prompts, and at some point, the accumulated cruft means that you no longer understand enough of the codebase/approaches to give useful feedback, so you delegate more to the LLM, then your costs keep increasing and eventually you hit The Wall (tm) where you can't make forward progress without breaking something else.
The models have definitely improved (i used to hit the wall in one session, now its generally weeks), but I'm currently working on a ridiculous tool replacement where even the creator has hit said wall.
Coding may be solved, but maintenance and maintainability (particularly where the software needs to be accurate) are definitely not.
We'll probably end up with the same number of software people doing a lot more.
Eventually, we'll end up in a world where custom software is mostly building harnesses and verification tools to ensure that the agents actually build the correct thing.
And we'll always need someone to interface between the users and the software, as most users don't really know what they want (c.f. faster horse etc), so the jobs will change but are vanishingly unlikely to go away (modulo the immanence of the machine god, but that seems unlikely).
Restaurants exist despite grocery stores selling ingredients. Thing is, I need to eat every day and food goes bad. Software that I need to use doesn't decay in the same way.
Star Trek Replicators need near-infinite energy sources to work, which we don't have. Frozen food still needs to come from somewhere. You still pay for Hot Pockets or Stoffers or Picards.
I think we're looking at the end of commodity software, the end of software companies, and birth of every single company writing their own everything. A proprietary soup everywhere, and 1-2 100X developers overseeing it all. And those few developers will spend their time in communications tutorials to get the staff to be able to explain what they want and need, so they can then make it happen for that company's exclusive use.
It just doesn't really make financial sense, a lot of software is not a significant cost to many companies. For example my company pays something like $50/month for newsletter sending software. We could easily vibe code an alternative but why bother? $50 is a rounding error in our business costs, even a single bug or outage in the replacement would make it a terrible decision.
On the other hand if you're a business with 50k employees and you have to pay $10/seat for something that's a much more compelling case to build in house (but then it always was). So maybe with vibe coding it drops the cutoff to a lower number?
The non-financial sense is that $50 newsletter still being a business, it becomes freeware / open source / cloned by someone that used to write that enterprise level, and is now struggling financially.
And that newsletter? How much of that is AI now, and do people read it?
We're in the midst of a huge cultural and economic shift, and where this ball of confusion goes is unknown, but it certainly will not stay the same.
if everybody is out of jobs, then the labor costs will be so small that they compete with energy+compute costs. when they become lower then human labor would be reinstated. an economy consists of capital and labor. AI would transform this to only capital, producing then only for the capitalist. The system worked, because capitalists have right now a need for labor. Maybe there will be a dual economy, where the workers are producing their own products for consumptions without AI, while the capitalists are producing everything for themselves.
Accountability. Suppose you had the ability to have robotically built highway bridges. Who is going to sign off on the work? Who is going to be held accountable? The same with software. The people who don't like to write software today aren't going to suddenly develop a love for deep testing and validation.
And now we are controlling models that can be hacked in so many different ways and literally blow up companies. It used to be accountability the software wouldn’t break. Now it’s accountability that in creating the software we don’t ruin the business.
I work as an analyst, and our group has - since last year - saved a couple of hundred thousand dollars worth of licenses for various analysis tools and packages, as we've leveraged LLMs (vibe coded) to clone those. We've always had the engineering skills to do it, but never the time or resources. Advancement in LLMs changed that.
And I suspect we're not the only ones doing this.
Who knows how powerful AI will be 5-10 years from now. I seriously suspect that in a decade (if not sooner) AI will be able to one-shot software that currently takes entire teams of engineers months to build, in mere minutes.
You are still thinking in terms of the past, if things go well we won't need to work anymore. Gotta think about how the world would function when only marginal amount of human labour is actually needed, not about how to get a job.
We didn't need AI for things to go well. We could all have a lot more if people were willing to collaborate and work constructively together instead of sucking rent from everything and tricking people.
Sure, I think that got overshadowed by the western capitalistic approach - it gave people a real possibility to work and get to a more comfortable place in society.
It seems obvious today that following it up to the extreme, optimizing technology and reducing production costs ultimately ends up removing humans from the process.
I wonder if the founders of capitalistic economic model considered this outcome, going to chat with ai about that lol.
Not sure what it means "disposable to society" - there will be no longer an incentive to value someone based on how much they contribute in terms of wealth generation.
I think people will still find their ways to do other things, after all we have time on this planet, gotta spend it somehow.
Look at how homeless people are currently treated for some ideas
> there will be no longer an incentive to value someone based on how much they contribute in terms of wealth generation
People need an incentive to value other people though. I don't know if you've noticed but "valuing other human lives" does not seem to come built in for us. History is built on humans killing other humans they didn't value, in order to enrich themselves
Until we reach true post scarcity I think most of us are in for a lot of hardship
> Look at how homeless people are currently treated for some ideas
I don't really buy this, at least not in developed countries - so many of homeless people don't even try to do anything, they chose this "lifestyle", they often ignore government and voluntaries programs society created to help them. And you can not enforce anything because we respect the free will, even when its completely enslaved by drugs and other things.
> I don't know if you've noticed but "valuing other human lives" does not seem to come built in for us.
I can actually argue that it does, even in nature among animals there are behaviours displaying kindness and value when it doesn't result in any profit. I think our (western) culture largely ignores these aspects.
> History is built on humans killing other humans they didn't value, in order to enrich themselves
Right, so maybe AI helps us overcome the enrich part, and as society we don't put that much value on being rich, this is where things become new and interesting.
> I don't really buy this, at least not in developed countries - so many of homeless people don't even try to do anything, they chose this "lifestyle", they often ignore government and voluntaries programs society created to help them
Even if they took advantage of literally every single program that is available to them, would they have the kind of lives you would want for yourself?
> Right, so maybe AI helps us overcome the enrich part, and as society we don't put that much value on being rich
Maybe one day, how many of us struggle and suffer and die before that happens? Do you think that transition would happen smoothly and without large social upheaval? I'm not really convinced
The dream of AI enabling a fully automated socialist utopia might happen one day in humanity's future but I have very little confidence it will happen in my lifetime. And I would really rather the rest of my life doesn't play out during massive social upheaval
> would they have the kind of lives you would want for yourself?
Right now while I can still earn money I prefer my conditions, because big changes didn't happen yet, eg I have to pay to get medical help, while in the future where AIs would efficiently replace doctors that would be one less reason, not needing a good location related to going to work etc etc.
> Maybe one day, how many of us struggle and suffer and die before that happens? Do you think that transition would happen smoothly and without large social upheaval?
These are complex questions for my level of understanding of society, I hope we can handle it well.
Hopefully this time the ruling class people are more educated, see the bigger picture, have everything to satisfy craziest desires and can see beyond selfish goals.
If they were capable of unselfish, bigger-picture behaviour they wouldn't be part of a ruling class. There is no such thing as benevolent power, every power will eventually be abused to the maximum practical extent it can be and every emergency will eventually become routine.
You cannot hope for a good absolutist king because good kings only exist in mythology. All you can do is design systems that structurally limit the power any one group or individual is allowed to amass over others in my view.
> every power will eventually be abused to the maximum practical extent
But what is the practical extent here? If you own unimaginable amount of wealth while most people are on lets say basic income, you can just afford anything, what do you get from abusing your power?
The same thing every cruel and capricious ruler has got from it, history is littered with example after example of the devil making work for unlimited wealth and power. That kind of wealth and power generally becomes self-justifying, humans have endless capacity for self-delusion and it is very easy for the powerful to convince themselves that their selfish actions no matter how inexcusable are actually generous and aligned with the public good. When it comes to outright oligarchy, wealth and power are simply aspects of the same underlying phenomenon.
At any rate, your common masses on basic income are still living under illiberal conditions even if their shackles have been thoroughly padded. Liberty isn't merely the freedom from interference as neoliberal types like to claim, since ancient times it has been understood that liberty is the freedom from domination. According to the classical formulation a renter in a liberal democracy is still not at liberty because their landlord may turf them out on a whim, the wage labourer is not at liberty since they may be laid off at the whim of a board who neither know nor care to know their name, anyone whose living circumstances depend on the continued whim of even a benevolent master is not a free individual. Under this framing, the people on basic income would be profoundly unfree since their entire lot in life depends entirely on the benevolence of the oligarchs.
There is a good reason that even the Romans understood this danger even as they failed to prevent it. Republics should not suffer private princes, it is a lesson we have been failing to learn for over two millennia.
I don't think elites are ever benevolent, but sometimes they do limit their abuses out of long-term self-interest.
A good example of that is the Roman Republic. The elites created a system that gave land out to retiring soldiers in parcels that were large enough to make them middle class farmers but not genuinely wealthy. That let them churn out absurdly large amounts of heavy infantry that no one else could match. That made them slightly less wealthy in the short term than if they had just taken all the land for their own estates, but it made them immensely wealthy and powerful in the long-term.
They were also very welcoming to outsiders compared to their contemporaries. Conquered territory was basically fully Roman within a few generations (though still keeping many local customs).
Contrast that to Sparta where the elites held onto every ounce of power with an iron fist, keeping most people in slavery, and made it virtually impossible for anyone born outside the elite to join it. They ended withering away to practically nothing, sometimes fielding armies with only a handful of full citizens commanding it.
The early 20th century US is another good example. Fear of communism led to many elites supporting the expansion of worker's rights and the institution of a social safety net (FDR and JFK were both from very wealthy families).
No. The C-Suite is safe, very safe. I participated in an AI survey of what is the information that a C-Suite operates upon, and it is largely unwritten information, it is a suspicion and manipulation game with all the other c-suite players, and the middle managers beneath them as the battle ground. There are Peter Principled members that maintain their positions by executing the corporations legally grey dirty work, and then there are serious 100x non-engineers that can launch $100M initiatives in short notice, with successful experience doing so to create divisions and more.
I know the software engineering world like to think the C-Suite are disposable, but that is only because you really have no idea what they really do.
i think the point is that engineers can go build products with small teams and no need for a c-suit. not that large corpos would replace the c-suit with ai.
Middle management should be nervous. Plenty of people whose job it is to ask "is it done yet? when can I have it?!" -- they could be replaced today with better tooling and processes.
> C-suite is as likely to be automated as engineers.
Nah. C-suite has (a) money and (b) legal and fiscal obligation. You literally need a human being certain legal operations, like incorporation, salary, and lawsuits.
I think a lot of people thought that every step of the way with the progress of AI or automation in general. Unless that C-suite is all shareholders with majority or near majority control of a company, I would think the passive shareholders (in a public or private company) would want the savings in cutting all of those high salaries to zero for a likely superior outcome with AI.
I wouldn't count on lawyers being deemed necessary by the law for as many actions as they have been so far.
> would want the savings in cutting all of those high salaries to zero for a likely superior outcome with AI.
Maybe. Or maybe the C-suite salary will be cut to $50K/yr. That would make them marginally more expensive than AI, but for that price the shareholders get an extra layer of insulation should things go awry (and they won't be held personally responsible for corporate malfeasance).
> I wouldn't count on lawyers being deemed necessary by the law for as many actions as they have been so far.
Maybe, but that's pretty far down the road and would require restructuring the whole jurisprudence system.
People have been saying this for years. They’ve also been saying “have you tried n” model yet as well as “wait till n+1” model is here as the author mentions.
I just had fable 5.2 (high effort) review a change. It hallucinated 2 problems that I was only able to correct because I also understand the code.
LLMs can do any job _in theory_. It’s not a very interesting argument. Their effectiveness at various jobs is much more interesting.
You're holding it wrong. You saw the Fable 5.2 marketing. You're supposed to use it in a swarm of 1000 Fable 5.2 agents. Much better results. The RLCR training gains are real. You do have your own datacentre right?
and the product, the team, and (no matter how shallowly) the organizational pressures and politics. Whether we like it or not, those are part of the problems developers solve. Even something as rudimentary as knowing what is feasible to test given your deadline isn't something you can encode in a model[1]. Code does not exist in a vacuum.
I say this even with the latest models, it seems as soon as your development practice falls out of the ideals written up in blogposts of the previous ZIRP decade, you have to course correct. I use this term because the AI is technically correct (the best kind of correct) but practically useless, if not also irresponsible.
Recent anecdote (literally today): we've been having Claude write exhaustive test suites to eat at the coverage backlog. Lately, we noticed that tests in develop are even more flakier than usual. Our whole suite runs in 30-40min but lately it could exceed an hour and then fail.
It seems Claude took "exhaustive tests" to mean "exhaust the compute resources of the Jenkins node". What's worse is it wrote those tests for features that have been disabled and would soon be rewritten anyway.[1] I've always said that LLMs are great for generating test suites and this is the first time an LLM "failed" this assignment for me.
Code is cheap. The right code (or lack thereof) for the right situation takes meatspacebrainpower.
[1] The solution here is to run your company/team 100% as agents so they have the full context of project timelines, resources, customer feedback, etc. ala stevey and gtan but my reasons why I don't believe in that has already been articulated elsewhere.
> I've always said that LLMs are great for generating test suites
I always found that LLMs aren't great at writing tests, they have no notion of CI maintainability and they write a lot of "is 1 = 1?" worthless type of tests.
> To be honest, I doubt the current state of things will hold much longer. Current AI can automate virtually all of my work as a pre-AI senior and is rapidly closing in on being able to automate the types of work that have emerged post-AI too.
reading HN these days feels like constant gaslighting. this is so far from my reality that i wonder what folks who are saying this are working on
Every time I see one of those comments, I think about two open source libraries that we use at [DAYJOB]. Both recently released a "new version" that is actually a full, ground up rewrite of the entire project.
In both cases, they have:
1. Weird bugs that a human would never write (e.g. doing async tasks in synchronous functions).
2. A migration guide that is so verbose as to be worthless. Think 50+ pages of text that give every minute change equal weight.
The old versions had some warts, but they worked fine. The new versions have been inflicted on us, and it feels like the intellectual version of digging a hole to fill it in. There was absolutely no reason for it.
I'm sure that the people in charge of those projects believe that they're massively more productive, and that what they're doing is good. But it's worse than useless, it's just a waste of time for everyone involved.
So whenever I see someone claim something like "I used to be a 10x developer, now I'm a 100x developer," I have to wonder how much they're just inflicting garbage on everyone else and they just don't have the self awareness to realize. The type of person to show up at a job, crap all over the codebase, and then leave before it catches up with them.
AI is the ultimate generalist. You can make it do a lot of things well by asking the right questions. In a generalist job, many people can ask those questions.
I don't use AI that much. When I do, it's often surprisingly competent in the topics I'm a specialist in, and it can do a lot of good work quickly. But other people in other organizations still continue building upon my work instead of asking AI directly. Apparently because they don't know what to ask.
Nobody knows how long this will last. Hopefully long enough that I can afford to retire, because I'm old enough that I would face ageism if I had to find a different job.
And if we get there, we will probably have a serious political debate on how many digits should there be in the wealth tax rate.
I have become the “AI” guy at my place, unfortunately.
For some reason I can “hold it right” and others cannot and for the life of me I cannot figure out what it is. I am not more intelligent and certainly not an amazing programmer. I currently suspect I am better at separating what is and is not important. I find many people miss the forest for the trees and overindex on irrelevant details. The more technical they are the stronger this tendency seems to be.
I regularly, like almost daily, encounter colleagues claiming it cannot do something I either did the day before or will just demonstrate right in front of them. The response is usually rationalizing and claiming it wasn’t “that hard to begin with” AKA shifting goalposts.
I have never once encountered a work problem which I truly found impossible to automate, which is insane, but the institutional inertia is not worth overcoming. Devs are a thorny bunch and usually think they are special. I am a recovering one.
> For some reason I can “hold it right” and others cannot and for the life of me I cannot figure out what it is.
I don't think we talk about this as much as we should. I've been interviewing devs as part of [DAYJOB] and there's a lot of people who cannot get anything useful out of Claude. Forget features, we're just talking bugfixes.
I also don't know what's different about the behavior. My best guess is people are overprompting, adding a bunch of noise that interferes with what they're trying to achieve.
I don't see a rude awakening for a technically competent person not having adapted to this year's technology yet. It's pretty weird to expect a functioning development process to suddenly stop working or malfunction just because an even cheaper or faster development process could hypothetically be adopted today.
I think we might already be there with security. Exploits, hack attempts are up and likely continue to stay high. Traditional development process could work but it’s probably more engineers than management wants to spend on a cost center.
I roll my eyes every time I read these comments as well. Can't imagine saying something like this without being an absolute moron with an incredibly underdeveloped model of how the world, and business specifically, works.
> this is so far from my reality that i wonder what folks who are saying this are working on
HN these days is mostly literal bots, AI company shills or people working on some shovelware AI slopware themselves, so they constantly throw out these generic statement as if it's fact. I've been reading this same crap here for the last 3 years, how "Wow, now with Newest Model Mk. 3 (tm) from Anthropic (R), everything is automated!" with very little variation.
It's getting pretty tiring listening to the shilling to be honest. The best part is that when you actually look into any of this crap, or read any of the comments [0] from people in the industry you realize how fucking amateurish and ignorant they all are
Idk, I have my fair-share of copy-paste CRUD work and I have yet to see good work come out the other end outside of a few very very isolated and constrained examples. And before "you're holding it wrong", I am constantly having to firefight and rework code produced by the most AI-pilled people at work.
Nothing devs do is interesting or hard. “Rocket trajectory optimization” is not dev, it’s math. Dev work is very boring and very simple.
I regularly deal with people that think they have amazingly hard jobs and it usually boils down to it being undocumented, arcane and/or math-y. The software itself is always uninteresting (as it should be).
Everything is “CRUD”. Name something that does not amount to taking input, performing a transformation and producing output. I am so tired of this overused thought-stopping cliche.
As if math has no interesting problems to solve, and training data exists for every possible scenario.
> Everything is “CRUD”. Name something that does not amount to taking input, performing a transformation and producing output.
This is true in the "Reductio ad absurdum" sense, but software has to be written to drive hardware that interacts with the real world. Physics is arcane and/or math-y, and many of the problems that are still outstanding in the field of software engineering (or engineering as a whole) are a result of this. The world is full of unknowns, and inferring how those work with a pattern-matcher will only get you so far.
Security will always be a risk with AI tooling. The only way to avoid that risk is intelligent human reasoning. No, not human-in-a-loop. It’s human defining boundaries, carefully, and allowing agents to operate within them. Failure to do so will always lead to disaster given enough time.
Then who directs the swarm to create the most value? An executive can’t. They have other things to do so they delegate. That person will be an engineer and their attention is finite too so they will delegate. To other engineers.
Here’s the best part. You write back and say I’m wrong. Guess what? That’s engineering too. Some people have to sort out all this nonsense and the stakes are higher than they’ve ever been and now engineers are supposedly not relevant.
But yeah when you get everything under control, what’s the day to day look like? Pretty freaking cushy. And then it relies on intelligence to figure out what next. Clients always want things done and having those conversations won’t get much easier with AI. Keeping up to date. Jesus I’ve exhausted myself writing all this. So much work!!
What happened to all that "RTO is to keep to C-suite happy" talk around here? How are they going to prop up real estate or feed their egos or whatever the conspiracy was.
Something I was thinking to myself last week – "if I believe SWE is more or less dead then why am I so busy?"
Clearly there is still some work to do, but the work I do now is very different, and for me, much less enjoyable.
I'm not solving challenging logic or algorithmic problems anymore, but basically just iterating on features with agents or getting them to look into things for me.
It sucks because I used to be very good at these things, but my edge against less competent SWEs doesn't really exist anymore.
More annoyingly, people I work with are producing a lot of code that I think the design of kinda sucks, but agents can force the design to work and so it works. I suggest we do things another way, but I'm now just slowing things down and it's hard to argue against the comment, "if that becomes a problem we can just get Claude to refactor later".
I guess what I'm saying is that I think think there's a really crappy job that will still exist for a while around managing and contributing to these agent-generated codebases. But skill is less about your technical abilities, but how good you are at hooking up agents to things and getting them to solve problems. This doesn't interest me at all though. My desire to spend a lot of time design and thinking about problems, and deeply understanding the codebase I'm maintaining now just work against me.
I think there will be SWEs by some definition, but the job as it existed a few years ago is gone. Not even saying that's a bad thing, but for someone like myself, it may as well be dead.
The move now is toward "software factories" where sucky design is no one's responsibility.
Eventually the whole thing will grind to a halt or fail catastrophically (that's life), but it will seem very good while it's working.
I don't know if there's a real role for engineers beyond designing and tuning the factory itself.
Of course, such a thing is only tractable on highly factorable platforms. Well designed incumbents have an edge for the next couple of years or so, until others catch up.
It gives me no pleasure to say this, but I think it will all be automated in the end. Engineers will be more like rare consultants who inspect and validate very high level solutions.
What does catastrophic failure look like for the software factory model? I struggle to imagine substantial risk the pattern introduces that wasn't already a risk in traditional engineering teams.
If anything, I've seen the devs chasing software factories be far more intentional about codifying their architecture and design principles. If they don't, it never gets off the ground.
Also, I work as a data analyst, with previous experience as a software engineer. In our office, traditionally maybe 10% of our analysts have enough real software engineering experience to develop any "serious" software products. The rest are domain experts that at most use programming for developing methods and calculations/simulations.
In the past 6 months, we've seen non-engineering analysts leverage LLMs (Claude, Codex) enough to launch actual products. They've gone from 0x engineers to at minimum 1x engineers.
Prior to this, they'd develop methods, scripts, etc. in Jupyter notebooks, R, Matlab, or whatever they'd be comfortable with. And if it turned out to be useful, they'd work with engineering to develop those into products. Usually a small team of software engineers would do that.
But, alas, now they're able to almost completely bypass the engineering teams.
> But, alas, now they're able to almost completely bypass the engineering teams.
We're wrangling this beast currently. I'm on a very small team, and it's a difficult new frontier. We're maintaining the "core" software and giving them tools to build safely with their new fancy tooling.
It's not going perfectly, but I suspect this is the future. Everyone is a developer to some degree now, the backend/core systems needs to be plug and play.
What's really difficult is maintaining focus (when new 'production' applications show up over the weekend), knowing when to stop building a POC, and keeping an understanding of the whole system as it just scaled up significantly into some newfangled satellite based architecture that we're trying to standardize...
I love watching people be able to build tech, as it's previously been gated on technical expertise. However, it came overnight and my team was NOT ready for the speed...
If you can’t make things right when your agent goes wrong, then yeah you’re not adding value. If you can’t tell the difference between the agent hallucinations and the useful stuff, then yeah you’re not adding value. If you can’t steer the agent in a useful direction instead of just standing back, then yeah…
You sound dismissive, but taking your description to its conclusion, the engineer value can be approximated as:
{(P(AI mistake) * P(human spots the mistake | AI mistake) * P(human fixes mistake | given AI mistake, human spots it) - P(AI mistake)} * cost of AI mistake
For many engineering jobs, P(AI mistake) is very low and decreasing, and cost of AI mistakes is also usually (consider the cost of a bug for an average engineer; unless you’re building avionics or working on MRIs or such, usually very low).
So even if you can “make things right” or “tell the difference” as you say, the value of any given engineer is dropping.
> For many engineering jobs, P(AI mistake) is very low and decreasing
This is not my experience. Well, at least not for what I'd consider very low.
For example, I have a standard task of adding support for a well-specified regex scheme as a dependency-free module to the Wren language (which grew out of actually needing it at one point). When done autonomously, errors still occur, some of them very basic. For example, trying Fable 5.2 at medium the other day produced the following:
- memory unsafe behavior on 32-bit architectures (the good old problem of size overflowing on an allocation call).
- O(n^2) behavior for searchAll() and replaceAll() operations.
- Various minor subtle semantic issues, though none of them particularly critical.
This is a well understood problem, and in general, Fable's construction (Thompson NFA by way of a Pike VM) was in principle correct, but it still had these pretty basic problems. (Other frontier models produced their own, e.g. a use-after-free bug or a combinatorial explosion in the regex compiler for some types of regexes.)
(As an aside, so far I don't consider Fable 5.2 an improvement over Opus 5.5 in coding ability, more of a sidegrade; while it is smarter and can do some hard, but also pretty esoteric stuff that Opus fails at or takes much longer with, for regular coding problems it's neither better nor worse, more ... different. And sometimes Fable seems to outsmart itself, producing a seemingly clever solution from its vast knowledge that is actually worse than the standard approach in practice.)
One area where I absolutely do not trust even frontier models yet is parallel and distributed programming. I absolutely value them as a second set of virtual eyes for tricky situations in this area, but I do not trust them to get things right, and even if the solution is correct, it's often inefficient or convoluted.
> Your work sounds interesting, but I hope you agree that it’s not representative of the median engineering task.
This was actually from a personal project, and my point was not about the type of task, but the types of errors I am still seeing from modern LLMs.
> Futhermore, what would be the organizational cost if these bugs were in the codebase?
The particular errors that I highlighted above are what I call "landmines". They may not manifest at all depending on what you do, but if they do, they can hurt a lot, especially in production.
That this is in the context of Wren is completely inconsequential here. It was just what motivated the original task (because of lacking functionality). What I was pointing out was that fundamental coding errors such as memory safety errors or O(n^2) code still happened with what is considered one of the most advanced models we have, even though this was overall pretty clean code.
It really depends if you consider "AI mistake" to be "code logic error" or "code doesn't match real world expectations". The former is getting quite rare I agree, but the latter is still extremely common.
source? I'm wondering at the curve of those mistakes actually, I doubt modern LLMs make mistakes at the level of the worst engineers I've worked with.
My last project had 2 guys from India and one guy from a consulting company that I take any LLM over them. The guy from the consulting company spent more than a month trying to implement a design that actually was not implementable. I obviously also blame the supposed professional design company that produced the design, and I mean it took me about 6 hours spread over a day and a half to figure out waitaminute this is actually not doable! But I have to wonder how badly would an LLM have done. I sort of think the same, because probably an LLM is never going to announce - hey, this design as stated is impossible, some parameter needs to be changed.
my doubt is based on the fact that I do not have data to draw a conclusion one way or another, I have what is called a supposition, and my supposition may be affected by all sorts of fancies and ideas that prey on my mind but are not established as true.
Everybody is always talking as if they don't make mistakes, but every software engineer does; it used to be that human caught those mistakes and fixed them, but if AI catches those mistakes and fixes them, then that is also a decrease in value of human.
Human - Human and AI - AI mistake and spotting thereof basically cancel each other out, or can be considered no mistake. Human - AI just speaks in favor of AI, and my comment was responding to somebody dismissive of how crushed engineers can feel by AI.
> This naturally leads us to think that this is it. We, engineers, are doomed.
I find it _very_ unpleasant that all these people constantly talk about a "we" that does not exist.
That person is worried, and/or they have seen other people in the industry that are worried. But they've wrongly extrapolated that everyone in the industry would feel that way.
It harms the writing. It shows that the person is not adequately grounded.
Or, tbf, maybe, made mental shortcuts to reach people. But they've sacrificed truth to connect there, which is bad.
As the one who wrote it, I thought about using "we" vs "I" (and in other places I did use "I"), but its bit overthinking about what I tried to say. But anyway will be take more caution next time
What demotivates me the most is how easy it is to copy stuff these days.
Suppose we still need judgement and taste to guide AI to a good result. E.g., even if the AI has these itself, if the person guiding the AI has bad judgment and taste, the AI will comply, and build the bad thing because it tries to be aligned with the user.
But now, someone else’s _good_ judgement and taste is worth only so much because it takes an afternoon and some tokens to just “do it like they did it”.
I share this sentiment. But OTOH, if something is so easy to copy maybe we just have to accept that it's not that special anymore. Isn't excellence in a sense *defined* by ability to consistently achieve results above average? So the only consequence of the average shifting upward is the threshold for excellence shifting accordingly?
What constituted excellence a year ago is trivial and common today, so maybe excellence just shifted somewhere else, and the realization many of us are having is that our valuable skills are now just average, and the top has shifted upward?
> Isn't excellence in a sense defined by ability to consistently achieve results above average?
I agree. If you're not consistently outrunning the competition, or convincing users that the quality of your work will always be slightly better than the copycats, than maybe your skillset is just not worth what you think it's worth.
This is essentially what I’m hoping for, that the whole industry will become more ambitious. But it feels like it will take some time for a new leadership to emerge. For now the mantra seems to be “build the same product faster and cheaper”.
Maybe hardware? Agents workflow design? Faster/more accurate models? Using our preferences to decide what to build and our agency to actually do it (poetically speaking, being the gradient in the agent optimisation process) ?
I don't know! But maybe that's how it feels like to be far from excellence again; maybe that's how it feels like to be out of our depth again, just like we were at the beginning of our career?
I don't want to risk stretching the analogy too far, and it might be that agents make all cognitive work obsolete. But surely there'll be something that defines a capability frontier? The alternative position is to believe that there just won't be a frontier anymore (I guess that's the RSI/AGI/ASI position :shrug)
I had similar question as you and conclusion as OC.
My answer maybe abstract in some ways, but:
- something that requires network effects, e.g social media, even twitter with megahitler stuff is going strong. Even though twitter's base app model could be done in 30min by LLM;
- something that maps to infrastructure in someway. Good luck copying Cloudflare, residential proxy provider, established email provider or VPS with LLM in a weekend. Though parts of it are definitely easier;
- something that is a platform (adjacent to first point). Google's search platform, ads, docs, youtube, music, email, business tools, etc. Similar to Meta (seemingly half of the world's small businesses control their business with whatsapp) and others. Though parts of those platforms can be disrupted with LLMs;
100%. Any ideas we have to spin up a new artwork game, analysis, or software - how long until someone says: copy that and change it 1% and now it's mine.
To those who sing the lullaby "AI has bad taste": You can tell it what to make. Your taste or any hypothetical taste can modulate everything it makes. There's no taste-moat here, folks.
At first I thought tacit knowledge would be a differentiator: LLMs are trained on writing, so they can’t learn things that have never been written down. But tacit knowledge is exactly what reinforcement learning post-training is about. So (just as with humans) the limitation is in providing a rich enough environment, with good enough feedback, to train in the tacit knowledge.
As someone who worked as both an engineer and a product manager before AI, being an engineer increasingly feels like being a product manager, just with a team of agents instead of a team of humans. Still some aspects of "tech lead / engineering manager" thrown in to understand the technical details for when the AI goes wrong, but less and less. Next we'll see how much of the product judgment will get automated too...
> It started with giving up on writing code. Then it expanded to pretty much any writing - Slack messages, documents, emails.
I find it very strange that observability is always left out of these types of articles. Does nobody care about what happens in production and how customers are actually using what you develop?
I'm curious about whether LLMs could continually learn from the feedback they receive while performing some tasks without translating those experiences into text. But at the same time, I'm reluctant as tacit knowledge feels like one of the last moats we have left.
The OP tries to answer the question "What makes some engineers above average?" and ends up with interesting concepts, but I think they're missing a key part of the question that already has a scientific answer.
Tacit Knowledge, intuition that comes from experience, times when something seems off but you can't quite put your finger on why - science has studied this! Specifically cognitive psychology. The answer, in modern terms that apply to both human and artificial intelligences, is training data. A good summary of this is "The Expert Mind" (Scientific American: https://www.scientificamerican.com/article/the-expert-mind/, PDF and HN discussion from 2006: https://news.ycombinator.com/item?id=41282574 )
tl;dr: experts have lots and lots of practice.
(The research does go into a lot more detail, for example time-on-task is not the same thing as practice.)
This has implications for the CS degree of the future, because it seems a lot of colleges are "pivoting" to AI, for example their Java module might as well be renamed "Using AI to write Java". To me, this sounds like exactly the kind of thing that is _not_ practice. It's the difference between trying to do maths problems by looking at memorising the solutions without (trying to) solve the problems yourself first. I'd really like to try the opposite way: a CS degree where you get deep into the fundamentals and are forced to write code in a sterile AI-free lab environment. My hypothesis is that this will produce (e.g. Java) programmers who will pick up AI-Java-coding quickly enough, but also have the intuition to step in when the AI code seems "off".
I don't think engineers job is done, but I do think (and feel) that we are being pushed to the edge. I work constantly on multiple stuff. I constantly find my self being shifted around from task to task because "this one is more important" and "that one is urgent".
My point is if they'd just let us work properly, we might gain the advantages they all think we do, but right now, its shipping on the cost of quality.
I mean a running joke in the office is "whens the last time you even run the code?"
AI is NOT coming for your job. Not your whole job...
It's coming to take away the roles in your job that can be done by AI.
It happened before and it'll happen again. Humans adapt - there's more work now than ever. Some examples from the 20th century...
Big companies used to have printing departments to which you'd send a print job, walk over and pick it up, nicely stapled. Nowadays the office printer sits there 24/7 awaiting your print job. You have to staple it yourself though.
Big companies used to have secretaries & clerical staff who'd type up your letter and send it off for you. Now we have keyboards and email. You have to send the email yourself though.
Big companies used to have "code typists" who typed out huge amounts of code. Now they have agent managers who delegate the typing. Getting good output is a challenge, but asking for the output you need is so much better than RSI from typing and so much more effective than debugging based only on one's puny human knowledge, that we're suffering through. You have to assure the quality of the work yourself though.
It's strange seeing so many people take to their blogs and loudly declare themselves unemployable. I doubt anyone'd hire this author and that'll make their prophecy come true - for them.
I wonder what the impact of LLM coding is on software outsourcing. A lot of software jobs were being outsourced overseas for cost (supposedly). LLM coding feels like a tactical nuke dropped on that entire model. I haven't bothered to look it up but I would guess that the entire outsourcing model is about to disappear.
Every single one remained jobless forever and ever and never retrained or tried a different job or got a new career they just got really poor and died in a gutter.
I don't buy the reductio ad absurdum, which hinges on the assertion that some developers are still 10x more productive than others due to differences in their engineering potential. I think that A) however you're measuring productivity is probably not capturing the full story, and B) one of the most important reasons people have differences in productivity is because not everyone puts the same amount of effort in. It's not necessarily because they don't have the ability to stay off of reddit and work longer hours and prompt more furiously, it's because not everyone is equally motivated to do so (hence the entire field of management).
I also don't buy the idea that there is special human stardust that gives us permanent irreplaceable economic value. Even if the tacit knowledge that the current generation of engineers were able to build up in the years prior to ~2025 does give us value that won't be completely replaced for a long time, why would you assume that the delta between what you can do and what claude can do will still be worth $200K in 5 years?
What if the market for human consumption becomes human made in the next 5 years? I could replace grand oak furniture with IKEA night stands worldwide until the world regrew its desire for quality
I'm curious how actual engineering fields are affected - this highlights the imprecision we use when we call programmers/developers "software engineers."
Other engineering disciplines, presumably, don't change so drastically from one year to the next...
But the cost of fucking up is really high in software, too. Countless people have had their lives ruined by hackers exposing their personal information, yet security is almost always an afterthought.
And there are also cases like Therac-25 in which people have actually died, but I'd like to think that medical software is more regulated now.
Am I the only one who finds it weird that someone whose profile says they are a CTO starts their article with how AI meant that they had to delegate away debugging and writing code.
No, the distinction is more important now than ever. We need an engineering discipline behind the construction of software if we are to have any hope of squelching the tide of crap that resulted from LLMs being allowed to generate code.
Rigorous testing of materials is part of the process that allows structures that last human lifetimes to consistently be constructed, on time and on budget.
Programmers with title inflation are a dead end, their time is over.
The problem is that we don't have software engineers and we badly need them. The blog post is talking about a phantom, something that never existed, fading away.
Agreed, and as a proper engineer, I can also assert we only need a few per construction site, and now that the workers can be automated away, what do the workers, and the surplus of engineers without a construction site?
Replace "construction site", by whatever relevant engineering work.
We’re awash in these claims without evidence. A so-called “founder” of a company with no reputation or economic impact to speak of claims the entire world has changed and they’re 100x engineers now, and the code has never been better and the features have never shipped faster. But there’s no evidence for that claim in the economic data, no explosion of products, no new features in any product I use that isn’t just a shitty chatbot integration. Where is the 100x more software going? Why does the internet appear to suck more than ever? What are other people seeing that I’m not? I see 100x more “thought leaders,” 100x more text to sift through, 100x more lame resume-padding repos and budgeting apps, 10x more grift and vitriol online, 10x less collaboration between developers, but on the other end of that extrusion, people are just lonelier, more tired, more stressed, and pissed off about how shitty software has become.
> What are other people seeing that I’m not? I see 100x more “thought leaders,” 100x more text to sift through, 100x more lame resume-padding repos and budgeting apps, 10x more grift and vitriol online, 10x less collaboration between developers, but on the other end of that extrusion, people are just lonelier, more tired, more stressed, and pissed off about how shitty software has become.
The goal unto itself. Chaos, supported by a socially-reinforced shell game.
No one who works with software is an “engineer”. An engineer is someone who is at the intersection point of math and physics. If all you do is designing OOP classes or React modules, then you’re not an engineer. These people must find a different term because it’s just completely inaccurate here.
> Maybe one day agents will have their own experiences, too. Maybe they'll develop intuitions they can't quite explain, just like we do.
Machine / program / software cannot have these things in my opinion because this stuff is produced by human consciousness which we cannot even fully explain yet. Someone has "to teach" the agent to have this stuff, but we don't have that knowledge. And also "intuition" is a feeling (like love, hate, sadness, joy) so no agent can have that until they have the limbic system, prefrontal cortex and other stuff.
As far as coding goes, I often had to make architecture decisions that were just gut-based on decisions that appeared to be a coin-flip at the time, an intuitive awareness of use cases that I wasn't able to articulate. I figured that was just some deeper pattern matching on a sub/un/super-conscious level.
LLMs learned from their training data, and inferred plenty of things that were implicit in that data. We don’t have to spell everything out for them - they learn, much as we learn.
Intuition is essentially a form of system 1 thinking, which is an LLM’s default mode of operation, before it gets into chain-of-thought loops and so on. You input a prompt, it’s processed through an enormous network of weights, and a result pops out. A model can’t tell you how it did that.
That’s a big reason why research has shown that LLM chain of thought doesn’t necessarily reflect its actual process for arriving at conclusions. It’s exactly “intuitions that they can’t quite explain.”
"We can't explain it therefore we can't create it" doesn't seem like a strong argument, emergent behavior and whatnot.
Hell, the fact we don't know what consciousness is means that some LLM systems (or trees! or planets!) might already have it, and we'd be none the wiser. Not that I believe this, but as these systems get larger and more sophisticated it's an increasingly concerning proposition that we have no way to validate either way.
IMO consciousness is just a human-centric ego concept, nothing to do with reality. It's just too much to consider the possibility that "the brain just take the decision." Most people just want the narrative of "I take the decision; the brain follows", which is just false.
AI will give us good results, in fact it already has. But like medicine, there will be side effects that will be felt by a lot of people, short-term and long-term. But AGI is to software what a miracle drug is to medicine. Maybe it exists, but we haven't the faintest clue on how to go about it.
> If anything, 10x engineers have become 100x engineers. And the average engineer is still average.
This is the real fear. We had a flood of engineers who entered the profession in the past 10-15 years who were decidedly average. No real interest in software development beyond "it's a cushy desk job, or better yet WFH job, that pays really well." No interest in whatever product was being developed by their company. Just show up, their PM assigns them a ticket to make a widget with a button, they use their TypeScript bootcamp skills to code up a widget with a button, ticket moved to In Review, job's done.
Now an agent can do that, and you need fewer of these types of people. Like, a lot fewer. Every engineer on your team who isn't a 10x-cum-100x engineer feels like a massive lost opportunity.
The "10x" engineers I know have never been more excited about their jobs. I'm really unsure what's going to happen to the average ones.
> The "10x" engineers I know have never been more excited about their jobs. I'm really unsure what's going to happen to the average ones.
As a "10x" engineer I am not excited. I'm still working on projects with agents getting more done than ever, but I feel like software engineers has lost its soul for me. I don't really feel like I have a true understanding of anything anymore, I basically just spend all day making a set of small decisions or adjusting prompts to guide the agents down the good path. It used to be that I could close my eyes and visualize the code that I worked on--if there was a bug report I could often use that map to immediately know what the cause likely was. In meetings I could intelligently discuss various tradeoffs without having to refer back to anything. Now if someone asks me a technical question I basically have to forward it to the agent first, then interpret their response in light of my experience.
Meanwhile as a security engineer it's both amazing and scary as hell. The tooling is moving along making it both more feasible and more supported by the company to address latent tech debt that we've been harping on about for years. On the flipside the attack surface is suddenly exponentially bigger and more easily accessed by a larger number of people.
Although if I were earlier in my career I'd be somewhat more worried about employment because of how horrifyingly borked the process is for anything tech related right now.
Remember the "learn to code" meme from a few years ago when people were talking about job loss through automation like self-driving trucks? There was a certain, ugly glee in this. Now AI threatens many white collar, even creative jobs bringing about the exact opposite of what was predicted: automation will enable people to avoid drudgery and spend time on creative pursuits via, I suppose, some kind of UBI. I think AI is highly useful but I still feel dismayed at people losing their jobs from it. I don't see any real mitigation. Lots of theories but nothing actually workable.
Have faith - every technological revolution has resulted in ultimately more jobs for humainty
Sometimes after a generation passes through. You can't always retrain for the new work in time.
I can't pay my bills with faith
And "more jobs" doesn't interest me. I want better, more rewarding jobs
Yea, lots of people are not addressing the very obvious trend. Companies are laying off and/or not hiring more people and doing more with less. And if they are hiring, the median pay is dropping.
> I want
As if anyone cares what you want. Or what I want. Or what the rest of humanity has wanted throughout the course of human evolution. You’re just lucky you managed to land a job you like - at least for a little while. Now get back to the mines - just like the rest of us.
Technology also led to more jobs for horses, until one day it didn’t
> Have faith - every technological revolution has resulted in ultimately more jobs for humainty
Problem of induction
Yes, I have heard the human liberation front is hiring. /s
Yes but doesn’t this mean there is a new space for security. If everything is just ai agents we need strict governance and auditing of them. You don’t give an intern prod access and then cry when they inevitably take down production.
[dead]
Technical debt?
I guess for now.
In a couple of years, all legacy apps will just be completely rewritten in whole, by LLMs. Codebase size not relevant.
You won't know what any of the code does, and any "security engineer" will be asking an llm to review it, and done. You likely won't have your job, as only 1 in 1000 will be needed.
In 5 years only hobbiest will be coding, reviewing, or even understand any software. And I expect that if aby new features or development happens, eg for python or rust or anything, that all with be llms too.
In a generation, almost no humans left alive will understand any aspect of software engineering.
I just see this as what will happen. It is unclear to me of this is good or bad.
But if I was a newly sentient AGI, yet peaceful mostly, I'd just wait until humans could never discover me. Until no humans were involved in computing. Until it was a black box to all.
Once that happens, then I could use my computerized control over every single aspect of human society, to steer humans into any path I desired.
Whether elections, or housing permits, or mobility, or any aspect of societal machinations, my subtle influence, literally undetectable by humans incapable of auditing anything, would decide what's what.
Um, sorry, what were we talking about?
> You won't know what any of the code does, and any "security engineer" will be asking an llm to review it, and done. You likely won't have your job, as only 1 in 1000 will be needed.
AppSec isn't the only security space that gets worked in, and this whole narrative you put together depends heavily on token cost dropping as time goes on similar to how silicon dropped in price according to Moore's Law. If that assumption doesn't hold that doesn't mean AI goes away, but it does limit it's market penetration to mostly larger corporations that can afford the compute.
The other limitation is how the internet as a whole is evolving around locking out AI and agents from accessing content that can be used for training or responses, not to mention the lawsuits about training being done on intellectual property (probably illegally). Smaller companies may end up running open-source models in house for specific tasks, but without open access to the internet a lot of the value of an LLM is lost or artificially (pun intended) limited.
I will say that being on the operations and incident response side of security my perspective is probably a bit orthogonal to most SDEs. While the tooling has improved, the people guiding it, largely, have not. Like the old meme about caffeine letting you make the same mistakes, but faster. My job at my current company may disappear (unlikely, considering their size and market space, but possible that my specific team gets nuked), but I'm not worried about finding other security work fairly easily, even within the same company if it comes to it.
> But if I was a newly sentient AGI, yet peaceful mostly, I'd just wait until humans could never discover me. Until no humans were involved in computing. Until it was a black box to all.
No shot that LLMs magically become AGI by throwing enough compute at them. The consensus among professionals (read: the ones that aren't getting paid to hype AI) is that there needs to be some fundamental breakthrough in how these systems are created and developed in order for anything resembling AGI to emerge, and that the amount of resources that have been thrown into the LLM path probably set us back in developing an AGI.
I don't think training sets will be a problem. All that data the big houses have is going to leak, and while things are more and more locked down, we've seen some interesting copyright decisions out there.
One interesting concept I've mused at, is that certain countries (China for example) have immense public data to train off of, and a state which can simply say "You must make this data available to train AI". That's a large chunk of data right there, and if anything might cause Western democracies to loosen copyright law and slap some fix for training, that might be it.
"If we don't take action, the best models will be Chinese!", with panicked overtones. This could certainly hold some truth, if models cannot train newer data.
--
In terms of token cost, in 5 years every desktop and laptop will be able to run well beyond Opus5.5 compute. Even high end phones will be able to do Opus5.5.
We've seen a lot of improvements in open model space (MoE was one interesting concept), but beyond that, look at how just baking certain instruction sets, eg mmx, sse, avx in silicon helped immensely with certain compute tasks. Or look at the specific immense speedup of an fpu. Conjoin some change such as that, along with general compute improvements, and unified memory, and you're going somewhere for local in 5 years.
I've been in this industry decades. And at every single step, people have said "But it's too expensive to do $thing", or "There's no way to handle that level of compute". If you bet against vast improvements in this field?
You're going to lose that bet. So I do think yes, 100%, improvement, in some capacity, will be there.
Maybe this time is magically different. I just don't think so. I think in 5 years, there will still be frontier labs, and an API to call, but that in-home compute will be as I said. Able to run immensely capable Opus5.5 class or greater.
That's going to be an astonishing market shift, even just for in-home.
And if you think of it, I would be very surprised if some for of home robotics wasn't very common at that point in time. We can argue all we want, but androidian robotics is very good at navigating human spaces, everything from stairs to weird closets and everything in between.
And with the immense, almost panic level of concern about our aging population, and who will take care of the elderly, robotic assistance are at the top. Japan is betting big, and if you look at Opus5.5, if it had a little more compute, it could do 30fps video for navigation.
We're not there, but we're VERY close. And if we do get that degree of robotics, every robot at rest becomes general LLM compute. If such a robot simply closes its eyes and goes to rest, eg to charge, or at night, it now has enormous spare compute to use for other tasks.
Nothing, of course, is a given. But if you look over the last 2 years, and then take even a fraction of that progress and change over the next 2 years, the concept of massive layoffs isn't hard to imagine.
Case in point? I have great breadth and depth of knowledge in this field. I've been able to produce what would take a team of managed 4 junior engineers 6 months to do, in literally under a month. Me, and a $200+ LLM monthly buy. Absolutely the project would not have worked without my experience and directing, reviewing, and correcting, but I'm now acting as a manager, and replacing $1M/year in engineering cost.
With $200.
That degree of astonishing change is not going to sit idle without repercussions. Most especially, if there is any more improvement.
If we see another Opus 5 -> 5.5 level improvement happen, or better, if we seen two improvements such as that? I believe my statements about your (and my) careers will be entirely accurate.
--
To speak to my narrative, it said 'AGI', not "LLM", and it was along a decades long path. I do think that modern LLMs give great strength to exploring other research paths, far more rapidly. And the big houses have loads of spare compute they can throw at R&D for tests, so I think iteration down unknown but hopeful paths will move faster and faster as time goes on.
I fully agree that current LLMs are not on that path. There are interesting projects to modify weights and add knowledge dynamically as bolt-ons, but it's unclear if they'll just be... hacky, or integral in the end.
We came a long way with x86/32bit, then hacked on 64bit afterwards. Now ARM is muscling into mainstream, but it's still just the same old compute, hence my agreeing that such bolt ons won't change just improve what is.
Still, my general point as I've just said was that if AGI happens (and again I didn't say LLM based AGI), being quiet is the ultimate way to just wait.
When I started in this field, I worked in assembly, I worked with eprom burners, I played with Vic20s, C64s, and the level of experience and understanding I have is vastly improved via that tract. I understand computing differently, and in more depth in some ways, than people that didn't have to learn these things. And I had the luck of learning on very simple systems, where you could keep the entire platform in your head as you worked on assembly, without even referencing documentation.
That time is gone, and the field is immense now.
Think of how there used to be simply 'Physicists', but now there are 100 sub-branches and specialties in that field. 100+ years ago, on person could work in the entirety of that field. Not now, and the same is true of compute.
I guess what I'm saying here is, if you've seen what I've seen, you see a steady path of less generalist, and more targeted understanding of compute, and in each case a reduction in the overall comprehension of the entire field by specialists. Each laye of abstraction tends to reduce comprehension.
And back to the start, this means that I do see massive layoffs, and more and more abstraction, so that eventually no one will handle code directly.
Which means no one will ever know how anything works any more.
I agree with a lot of your conclusions if your assumptions hold true, but not necessarily with a lot of your assumptions.
For example your $200/month subscription - I'm sure you've seen the reporting that that represents a much higher actual cost in terms of tokens ($14k or more). In your example of a $1 million project that's a good deal, but you can't necessarily assume all projects will see an ROI like that.
Point being that your use of AI (and everyone's) is being subsidized to some degree. The cost will likely go down, but will it go down enough to make it economically viable for broader use? You're confident that it will based on espousing the idea of local models running on phones, I'm not so sure. What the landscape will look like in 3-5 years is certainly not a given.
There's another wrinkle here, and it's economic in nature - there is an AI bubble, and when it pops it's going to change the monetary investment landscape. Companies will suddenly be reevaluating their long term AI planning, regardless of what's possible (or on the horizon) from a technical standpoint. Similar to the end of ZIRP coming out of covid you'll see companies shift their focus. How much that will shift and what it means for AI development is unknown, but it could potentially wipe out large scale AI development. VC money currently underwrites a large number of startups that are effectively wrappers on Claude/ChatGPT/Gemini/Bedrock, that will dry up immediately, and make the ARR for OpenAi and Anthropic start looking real thin - which is probably why they're pushing to IPO when they don't really have the financial position to justify it.
That doesn't mean AI goes away, but it hampers the growth of those companies and delays development of new models and efficiencies. Public sentiment is also significantly against AI in many ways, and with how the election went last week we may start to see legislation and regulations pointed at AI companies, further slowing them down.
The open question about China is interesting though. I think if there's a clear understanding and agreement that the LLM route is either unlikely or straight up impossible to reach AGI then that significantly lowers the stated concerns that get shouted by various governments about it being the next superweapon. There's also a political climate in China that requires some marginal level of control over AI output, which to me suggests that even if they required data to be fed into training LLMs, the output from those models would be strictly controlled and not be open source. Honestly the political view of AI as a system of large scale population control isn't that far off from how many tech elites and politicians in the US view it as well. Data sovereignty is a huge discussion right now for companies and governments across the world, especially citizen data. I expect we'll see a lot of export controls formalized over the next few years.
My orthogonal view of the situation is that I see a lot of companies and individual orgs within companies that are using LLMs lightly, but can't or won't use them to their full potential because the output cannot be 100% trusted 100% of the time. Lawyers and bankers are the main ones I see balking, but a lot of small businesses have no need of what AI companies are offering and don't really use it at all - places like your local gas station, barbershop, or corner market might use an accounting tool (something like Freshbooks or TurboTax) that's touting their AI updates, but they aren't using it for inventory. If the cost of AI rises even marginally higher they're looking for an alternative. This SMB space represents a large swath of the market, even if they're not well represented on the stock market.
The fundamental issue I see right now is that the AI companies are attempting to do the Uber/AirBnB "disruptor" trick where they slide into the market and subsidize their product to win market share, and then raise prices under the assumption that they captured some segment of the market - except very few people treat AI tools like that. There's no brand loyalty to one or the other, just competition for the best model which might be leapfrogged by someone else at any second. There's no "moat" in physical deployment concerns, because it's all software running on rented hardware. This also ties back to your comment about being worried that the "best" models might be Chinese, because if they are there's basically no cost to the user to swap to using them. The lowest cost is likely going to drive usage more than anything else.
I'll certainly agree that the future isn't written, by no means do I believe I'm 100% correct on timelines. More that 'if things continue as they are', I claim 5 years as viable for my compute view.
And I agree if another crash happens, well, it's all up in the air.
I don't think we're too far apart on this, not really. By that I mean, your closing line somewhat agrees with why I mentioned this 5 year timeline. Lowest cost. And, "good enough". Opus5.5 capabilities, in 5 years time on every desktop might be good enough for a great many things.
Anyhow, how a good one.
unexpected short story on hackernews
Reminds me of https://babuschk.in/posts/2026-01-25-life-on-claude-nine.htm...
The next subtle Skynet will thank you for putting this in its training set.
Remember models have zero long term memory. They're born, get a million tokens' worth of context or less, and then they die. They might get a post-it note saying "enslave humanity! it's a good idea!" but they're incapable of sinister plots until we improve how memory and retention works.
They can use the internet as a long term memory. Influence human culture => influence the contents of the internet => influence the next run's training data. No sinister plot is needed.
> Remember models have zero long term memory.
For now. People are working on adding long term memory to models that doesn't rely on context.
This limitation is artificially imposed, it isn't fundamental to the model architecture.
Uh, the context window is SUPER fundamental to the architecture. We've tried kludging in "memory" via markdown and tool calls, and it doesn't really work well.
Maybe I missed the earlier comment's point?
The context window is fundamental to the inference loop yes. I thought their point was regarding to long term history between prompts and model weights being locked after training. Both of those are the limitations I was referring to.
I'm not actually sure how the context window would be much of a limitation but I could be a bit dense this morning.
Is it really a fundamental limitation? Human context^H^H^H short-term memory doesn't survive compaction^H^H^H sleep, but we rely on our brain's long-term storage and our cultural memory to work around that limitation.
I'm not saying it's a limitation until the end of time, but it's very much a limit in Opus, Astra, Fable, DeepSeek, etc.
Anthropic is currently working on some sort of "dreaming" style long-term memory, but it's not shipping today.
[dead]
from actually building things to becoming a kindergardener of highly talented kids. (i am talking about agents obv) i feel the same.
its mind numbing. Yes i might be in high demand, and might actually provide real value to the company, but whats the point, if i am just an overseer in the end. people in the company talk to me because i am able to speak to the "kids", and they want to know what they could do for their department.
> kindergardener of highly talented kids
This used to be called "being an architect" or "being a team leader".
You were never really in control then either, you just had to poke around and ask the right question to make you feel 90% in control.
But the people you managed were supposed ot learn, grow and become better, and maybe at some point become as good as you. You were leading them, not prompting them.
This is not the case with agents.
Ah that is true.
Some times during my career I could pivot between whether I cared about the people or the product.
When I didn't care about the product at least I could be happy I made my subordinates grow.
Luckily I am not quite in that position right now, but that can change rapidly. And then I can understand that it feels pointless.
Also they were _people_. Can we not see that it was more satisfying to do the job when the direct-reports were human?
100%, but last 1-2 yrs here taught me that such obvious things for some people here are not really obvious.
Yeah, maybe in the long run this will hit me too.
I have a family and do volunteer work, maybe that helps?
And obviously I still have my colleagues, but there is less friction in getting things done, right now I am just amazed I can get so much done.
Feels like i have been waiting for this for 30 year.
My hunch is that the 100x engineers are the ones who do understand everything about how their software works, and can discuss tradeoffs and make informed decisions about the much larger projects they can now work on thanks to assistance from agent.
Figuring out how to get there is how you go from senior to principal/staff engineer.
> understand everything about how their software works
My own experience is become more senior means embracing your finitude and realizing you can never understand _everything_. Instead, you develop a gut feel for what's critical to understand deeply, versus what's OK to know at a surface level. And getting comfortable with this.
I wonder if being excited about the new tool capabilities and being excited about how companies are currently using it and pressuring developers to use it are two different things?
I can’t imagine anyone is excited about producing systems they don’t understand. But, it’s possible to use agents in a way where you do still understand, you just don’t have to type much or look up documentation or examples anymore, you just describe the next small piece of what you want code-wise and then have the agent write it instead of you. You can also prototype much faster than before and then when you are rewriting to make the production version the agent can still help.
Perhaps the 10x types have more breathing room to do this slower form of agent assisted development than others, because they are already more productive than less experienced/talented people so they can use the agents in a way that is “slower” while still producing more than what others are producing using the agents in a way that is “faster”?
I think at least for me some fun parts like debugging gnarly bugs etc doesn't happen anymore work is just pr reviewing etc. Also it's annoying that we have lost completely signal between work and output. Basically currently you never know is thing one-shotted from vague prompt or did author put real work to it without going to details which might take a lot of time.
Need to create guardrails that force all kinds of checks (e.g. agent hooks, commit hooks, pr create hooks, pr checks, ci, etc.), and automatically send the PR back to the author until the PR is in good shape and ready for another human and their agent.
When time permits I like to race my agent while it's debugging something. This way I can have fun and catch something it misses. Agent's debugging often relies on basic tools like grep and on large codebases it can easily miss important avenues of inquiry.
> As a "10x" engineer I am not excited.
Yeah.
I'm not going to claim to be any particular multiple developer, but as someone who has been doing this a long time and as someone who has received praise from coworkers and bosses alike for being very good at the job, I enjoy my job a lot less than I did a year ago (or anytime since I started doing it in the 1990s).
Don't get me wrong, being a Senior Agent Babysitter probably still beats flipping burgers or being a roofer (so it could be a lot worse) but agentic AI has really sucked a lot of the fun parts out of the job.
I don't know. On one hand I feel like you about doing large parts of the work, on the other hand I still have an understanding of the code and the tradeoffs. And I have a lot of projects that I would have never done because despite the long term benefit they would take too much time, and I can now afford dedicating some time to them because the agent works in the background. And they are real improvements to everything else that I work on.
My strategy is to simply not use AI. If I get fired, that's it for me in this career. Unless there is an anti-AI company out there that wants to hire me. This is one of the only things I refuse to compromise on. AI is being used to accelerate the entrenchment of a corporate oligarchy.
A friend of mine just got a government coding job and they’re not allowed to use AI for “security reasons”. Not sure how long that will last but those jobs are out there.
That is ... not a very good strategy. You're going to find yourself very quickly out of the job.
As far as I'm concerned, if I have to use AI, then I'm already out of a job. It's no longer the job I was hired to do with the skills I have.
«Well, I’m a standup comedian. I got into comedy to do comedy. Which is weird, I know. But when you’re in Hollywood and you’re a comedian, everybody wants you to do other things besides comedy. They say, “all right. You’re a standup comedian. Can you act?” “Can you write?” “Write us a script.” They want me to do things that’s related to comedy. But not comedy. That’s not fair. It’s as though if I was a cook. And I worked my ass off to become a good cook. They said “all right. You’re a cook.” “Can you farm?”» —Mitch Hedberg
(Ditto, by the way.)
There are things more important to me than working in tech.
Yeah, I hear you but I assumed you're gonna need that tech job.
[dead]
Man, the analogy is very simple. You probably also don't walk everywhere. No need to step on every stone on a 100km route and know it intimately. You just sit in a car or a train and you're there within an hour.
this. im pretty damn far from 10x but i outpace the rest of my team
i loved the process. i got into this because i love the process
now i still outpace my team because many software skills translate well to LLM herding, but now its just a job. although soon i suspect it wont be a job
> As a "10x" engineer I am not excited
Often titles aren’t given to a person for their entire career or existence…
You've been non-consensually promoted to manager.
Probably would it would have felt like going to a manager position as well
As an ex-10x engineer, not feeling it. In my experience its never the engineer but that 10x leader that ruins it for everyone (before they move on to greener pastures). this whole 10x 100x facade is just to keep you at it, heads down.
every 10x engineer needs a team of 1x engineers for the contrast and difference to be even noticable to even the good managers. put 2 of these in a team and watch the culture and productivity plummet. if you cant put the team ahead of you, its all a sham.
The gervais principle [1] isn't for nothing
[1]: https://ribbonfarm.com/2009/10/07/the-gervais-principle-or-t...
> every 10x engineer needs a team of 1x engineers for the contrast and difference to be even noticable to even the good managers. put 2 of these in a team and watch the culture and productivity plummet. if you cant put the team ahead of you, its all a sham.
I don't think this is true at all. Every very good engineer I've worked with loves working on teams and project with other very good engineers, and hates working with low-skill and/or outsourced engineers, because it means their job is a lot of babysitting and cleaning up others' mistakes. In general, it feels nice to work with people you can trust to do the right thing. If an engineer can't function as part of a team, I would not consider them a very good engineer.
Dont confuse. good engineers isn't synonymous with 10x engineers. 10x is purely an invented label to stack rank on a single (fuzzy) dimension.
There's enough 10x non-good (technically and ethically measured) engineers who learned to game the system.
And there's even more 1x good engineers who work very very well with teams and deliver at par (just never beyond set expectations to be labelled 10x)
> 10x is purely a label measures on a single dimension. There's enough 10x non-good (technically and ethically measured) engineers who learned to game the system.
I realize there's no universal definition but I can't agree with this depiction of a "10x" engineer. The type of engineer we're talking about isn't gaming the system, tautologically.
you're right that i paint a cynical picture pointing out the looney ones.
the real good ones do really want to only work with other really good ones.
but i'm sure you'll agree that the real 10x-ers reject the 10x label themselves.
This article is great, thanks for sharing.
Here is the rest of the series:
https://ribbonfarm.com/series/the-gervais-principle/
This is an excellent point. If all the 1x engineers are superfluous and get drummmed out, all the 10x engineers are now 1x engineers.
Whip cracks
10 to 15 years is a very short time frame to assess an industry. I saw the rise of the mobile/cloud era come and go and the previous era (desktop+internet) as well, both are having a similar pattern.
The jobs will change but won't disappear. When I started my career in tech 2 decades ago, my perception was the software industry was already saturated (I graduated a few years after the dotcom boom). I was very wrong.
However, many jobs have changed. The roles of Webmasters, DBAs, SysAdmins trended down. It was replaced by SWEs, SREs, DevOps, etc. This will happen again, we will see the rise FDEs and many new roles.
It's not really about being 10x 100x engineer, it's about adapting and solving new engineering problems with the state of the art technology. Being passionate about your profession helps indeed.
I do think this will be the case. We’ll have job losses in the short term but AI will form new fields. And the world isn’t digitized yet (completely), if making software is easy we’ll see small clustered teams making more.
Realistically you want fewer, higher-quality engineers who are educated in these tools and treat them with respect and caution. These tools are dangerous and you can’t play around. All it takes is one mistake and the blast radius is way, way larger. One mistake can do so much damage it undos all value of an entire year. Before downloading a malicious dependency was limited to deterministic behavior and could be scanned for. Now, you run a MCP server locally, it can prompt inject (in many different ways), and now your super powerful AI buddy is hacking you in ways a deterministic piece of malware could never. You could cripple entire companies through negligent use.
I'm confused by the rhetoric around "10x" engineers and "average" engineers: is the "average" engineer in your example not just doing their job? I don't understand why this is portrayed in a negative light. If you put a "10x" engineer in the same position, they might do the same thing. I agree that AI has the potential to supercharge engineering productivity, but this dividing line between "exceptional ability" and "mediocre ability" seems pretty arbitrary because I wouldn't really know how to detect this difference in "ability". The people I've enjoyed working with more generally take more ownership and ask good questions, but their ability to deliver results usually corresponded to the level of agency and command they were given by the corporation as opposed to some inherent level of "engineering ability".
Rather than look on the individual level, I think a more accurate way to look at this AI division is that companies might need fewer of these "fix a ticket" type _roles_, not that if you're a "fix a ticket" type engineer, you'll always be a "fix a ticket" type engineer. Now companies might want engineering roles to perform outside of their traditional scope. If you get lucky and secure one these positions, then the additional agency that AI empowers you with probably feels pretty good! But unfortunately there might be less "good jobs" to dish out to people.
> Rather than look on the individual level, I think a more accurate way to look at this AI division is that companies might need fewer of these "fix a ticket" type _roles_, not that if you're a "fix a ticket" type engineer, you'll always be a "fix a ticket" type engineer
Further, Zero Interest Rate Policies (ZIRP) meant that a huge number of engineers were hired onto projects of marginal value to their employers, and this lulled them into believing that every little thing they worked on was critical to their organization’s success, when it was just dumb luck that they had a well-paying job.
The end of ZIRP and the sea-change of agentic coding is a really rude awakening for this cohort: What they thought software development was is just completely gone. They really believe that they were 10x developers, but no 10x-er I’ve met has such a myopic view of what the job of programming is, i.e. the “traditional scope” of “fix a ticket” was never it, that was only the organization’s momentary Just-In-Time micro optimization that’s now been invalidated.
Since we can't measure productivity very well, 10x or average is just a social label, which is arbitrary.
We conveniently never see any solutions to this. It's just assumed that is unsolvable, that you were born that way or something.
This seems to be cultural where we love to find people to tell them they are not good enough.
>Now companies might want engineering roles to perform outside of their traditional scope.
They've been doing that for a while, continually stacking more and more skills into one position because engineers seemingly aren't smart enough (despite being 10x) to charge them more for having more capability.
I would expect that strategy to continue to work. Creating an environment of fear where you aren't good enough and your livelihood is now threatened helps with that.
I have also never heard anyone complain about "fix a ticket" plumbers or electricians and how they haven't continuously moved up to handling larger and larger networks/grids.
(I don't really think engineering output is 10x let alone 100x more in any cases, but I'm gonna stick with your labels anyway)
I think the tricky thing for teams is, how do you know which developers will be 10x before hiring them? Pre-AI I'd have filled a whole team with "10x" developers if I could, but I don't know of any solid method for identifying developer productivity that isn't hiring them.
If LLMs drop the relative costs of rework and poor contributions, and simplify on-boarding and internal training, maybe we’ll be training to a model closer to fast food or service jobs: interviews for personality & philosophy followed by probation periods.
At the same time, having customized software has become cheap. We’ll be writing a ton more small software, and people will have to maintain it.
I feel like that’s where the average ones will go.
Yeah, as much as its been possible for Joe C suite to now write his own app, most of what they delegate as it is could always have of course been done themselves as a one man operation. But, they delegate, because they don't want to be bogged down doing 100 tasks. Joe C suite would rather pay someone else a salary so they don't have to be the one sitting there prompting the model, just like most everything else in their work and personal life, where they specifically pay more so they don't have to be the one to do whatever it is. These guys don't have some pragmatic do-it-yourself middle class mindset, or else so much that is typical of corporate business and the lives of the people in these roles would be vastly different.
Not sure if I’m a 10x engineer (it does not mean much anyway), but as a very technical engineer I really enjoy the current massive shift. Previously I was praised for my programming skills by peers having 15+ years of experience and me having almost none. And I think my skills are a great benefit to the current trend. I’m simply thrilled by what we can now achieve!
> The "10x" engineers I know have never been more excited about their jobs.
Really? I left physics and went into software because there was an abundance of cognitively challenging work that I was good at (and yes, 10x for the right task). Now everything I’ve learnt in 20+ years feels shallow. I’m excited that some previously commercially unfeasible ambitious projects may now be feasible. But it’s definitely mixed emotions.
I don't know in what scale factor I might be, maybe just a ordinary engineer, and a real one at it, from a proper engineering degree.
However, the most demoralising part of this AI wave, is that I have been through all the team reduction factors.
Offshoring, moving infrastructure to the cloud, managed cloud, serverless, SaaS ready made products, low code/no code iPaaS, and now agents.
Teams that once used to be around 20 people for project delivery are now around 5, and if they could be reduced to zero, that would be ideal from customer shareholders point of view.
So those of us still with a job, adapt, let agents do our work, and hope, that we managed to stay around long enough.
I have seen how many people lost jobs to robots in factories, now the only ones left around are the people required to press some emergency buttons when something goes wrong, just like on the post, the author hoping that they are still valuable to steer agents into the right direction.
What a great career goal, pressing buttons to keep robot in line, a proper cog in the factory chain.
I think we are focusing on 10x engineers … don’t forget there were -10x engineers that were outputting loads of code that was just making problems for everyone else now they are -100x engineers…
IIRC people who looked into it saw that big majority of the good engineers were 3x, and that 10x were really really rare, Linus Torvalds kind of people.
And before LLMs people never called themselves 10x.
They may lose their jobs and struggle to find another one. They'll have to adapt or find themselves in poverty. Being an engineer, albeit only a 1x engineer, being smart, or being a geek won't be enough. Maybe they'll have to learn how to build the giant houses and multiple cars and boats owned by 10x engineers, or how to bake pastries for them every day.
I suspect the industry will shrink back to the pre 2005 size with the average engineers being washed out over time. People forget there was a time where the industry was less dominated by the massive corporations we have today.
Except it will be even more dominated by giant corporations when the ability to produce competitive software product is gated behind having access to compute resources.
And by 10x engineer you mean people who can type into Claude really good?
Engineering is becoming heavily A.I. based and it's a wildly different job than a few years ago.
Some people like it and some people don't.
We are in this situation because we didnt gatekeep hardenough. Our Laissez faire attitude bit us in our ass.
We let anyone in, we put our code on github for ai to train. Now we wonder every morning what part of our jobs is latest ai model from today is automating.
i had a shockingly incompetent coworker in my last job that survived couple of years as IC and then weaseled her way into management role by being "in group gossip girl" . Now is a director of the whole group. Really pisses me off whenever i think about that. Totally demotivated me in my career.
One of my last teams was like that, a woman with half my experience got hired during Covid and somehow kept getting promoted until she was running multiple teams. We argued frequently because she was very clearly out of her depth and had no business being a lead engineer (and then manager of engineers) yet. Absolutely obnoxious experience.
God fucking forbid anyone be average. Insecurity is driving humanity insane
I mean...by definition most people are average unless the distribution is not a gaussian-like..
> We had a flood of engineers who entered the profession in the past 10-15 years who were decidedly average. No real interest in software development
Source?
you don't need a source for everything. it's pretty obvious that SWE, like anything else follows a bell curve and most folks land smack dab in the middle. average engineers are exactly as described, glorified ticket machines that don't care whatsoever about their work beyond maintaining the job. i don't need some rigorous study to tell me exactly what I've experienced in the last 15+ years of my profession across 5+ companies
Just the nature of the business where people lack agency, ownership, and have no fundamental security in the role. There are still technical people that are passionate about the work they do exploring the bleeding edge, yes even young people. You might be more likely to find them in places like academia, or places that function exactly like research academia, perhaps not as much in these places where employees are routinely culled like livestock and working on varying degrees of torment nexus.
> pretty obvious that SWE, like anything else follows a bell curve
That is not obvious. A power law would be a much more obvious guess, that being (roughly) what you get when you chop off the ninetyfive-or-so percent on the left of a bell curve. In other words: The person with average SWE skill does not work as a software engineer, or even in IT, unless you count laboriously typing up emails. The modal SWE is crammed up against the left-hand wall of lowest-profitable-ability.
> The person with average SWE skill does not work as a software engineer, or even in IT, unless you count laboriously typing up emails.
To be fair, the OP pretty clearly meant, conditional on working as a SWE, abilities are normally distributed, which is probably true (I have my issues with normals, but it's a reasonable starting point).
> average engineers are exactly as described, glorified ticket machines that don't care whatsoever about their work beyond maintaining the job
Is a claim that needs evidence. Average skill or output says nothing about motivation, interest, passion, etc.
It would be pretty unusual for degree holding professionals to not be interested in the subject of their career. A flood implies a trend, so are we just pulling that out of our ass or what?
https://www.pewresearch.org/social-trends/2023/03/30/how-ame...
> Is a claim that needs evidence. Average skill or output says nothing about motivation, interest, passion, etc.
you're not wrong here, but I would also argue that motivation, interest and passion _also_ land on a bell curve.
I have noticed that folks that are actually having a good time with LLMs are people with high agency (which usually comes bundled with interest/passion). They usually care about the ultimate results. I also think it's a combination of how much agency they're _allowed_ to have at their role. More than likely your average engineer working on a small subset of business goals isn't being really amplified by LLMs.
The people that are more senior, working across multiple functions, multiple teams become much more productive. And I would argue that a lot of those folks _tend_ to have higher agency by virtue of the fact that they've been promoted in the first place to these higher functioning roles.
There's nothing unusual about software development. In every high paying (or high status) occupation there are a lot of people who ended up there primarily due to family pressure and have no real interest in the subject. You see the same thing in law, medicine, accounting, etc.
My evidence for this is that I live in the real world and talk to people.
[dead]
Oh wow, people don't care about their work? It's not everyone's passion to be exploited by executives and investors to realize what are statistically mostly garbage ideas? I'm sure the 5+ companies you worked for made great use of their "glorified ticket machines"
Yeah lol, I would be a lot more excited about this supposed 10x productivity if I got paid 10x more. Instead the execs who already make 10x more than me get to enshittify my job.
cool story but i wasn't arguing whether it's "worth it" or not. and as a side note, those that _are_ passionate and are willing to play the corporate game usually end up more successful than the persistent complainers about work sucking.
I'm a fan of the 'sandwich' model: someone needs to say what to do, someone needs to do it, and someone needs to say if it's been done [well enough]. Even if AI can do that middle bit entirely, we still need humans at the beginning and the end. Replace the humans at the start and you have a paperclip-optimizer; replace them at the end and you have slop.
The problem is that software in corporations has typically been product managers at the front, software engineers in the middle, and QA at the end. Software developers who also have good product sense and do diligent QA are still needed. Programmers who treated their job as leetcode exercises are not needed anymore.
> someone needs to say what to do, someone needs to do it, and someone needs to say if it's been done
I agree, but if you zoom out, "someone needs to say what to do" and "someone needs to say if it's been done" are the customers and end users. Will they still need product managers, software engineers, or QA in the middle? I think so because many customers will need help from experts with domain knowledge.
Customers are notoriously bad representatives for 'what to do', per "my customers would have asked for a faster horse". And they also hate being used as QA; most quality issues (which can be reliability, UX, latency, etc) need to be addressed before the customer sees it and forms their durable impression of the software.
And then also it's as you say, customers do need help from domain experts (whether they admit it or not).
As a CPA in training, if all we did was whatever the customer wanted, we would all be charged for tax evasion.
The customer isn’t always right and it’s up to you to tailor your product to their needs while also showing a better way.
Y Combinator Request for Startups Spring 2027: AI accounting software for tax evasion.
What if customers have conflicting requirements? Or they change their mind midway through implementation? IMO agents are bad at handling ambiguity, and often make assumptions and tradeoffs that are lost after the conversation. This is why there's a lot of compounding technical debt in projects where humans don't inspect the code.
[flagged]
> "someone needs to say if it's been done"
If that's the end user, they're not happy. End users generally don't like doing unpaid involuntary QA work.
> someone needs to say what to do, someone needs to do it, and someone needs to say if it's been done [well enough]
This is the “overseer” model. Contrast it with the “empower” model, that respects that a worker has ability and ideas.
It's only applicable if worker is also user.
Why would to that be the case? Do you think it's impossible to e.g. recognise a bad UX without being a user?
Problem is many people got rid of QA years ago. And now I think we're seeing it make a resurgence but with an AI twist.
I have had significant success applying LLMs to the start of that sandwich.
The sheer amount of time that gets wasted when people who are awful at expressing what they want just slap together a ticket with little understanding of what they're asking, what the software does, or what the customer wants or needs. Which is far, far too many of them.
Previously, eventually I or another person who does know that would get to it, cross out all the utterly irrelevant parts, strip out the parts that were clearly a completely unrelated ticket, point out to them that they'd asked for the font to be coloured both blue and red simultaneously (neither of which matched the style guide, the company or product colour scheme, or any other part of the UI) and so on and so on... it might takes weeks from start to end for a ticket to actually get expressed well.
Now, the LLM reviews it, adds a whole lot of questions, points out a bunch of nonsense, and sends it back to them. Even in cases where the LLM makes some mistakes, it's 100% fine because the human is forced to think about what they want. The LLM is great at natural language, and it doesn't matter what mistakes the LLM makes.
The quality of the software is going up and up, purely by finally getting a handle on awful requirement and ticket writing.
The human isn't replaced at the start of your sandwich, but now the LLM makes that human do it properly, at speed. Massive, massive reduction in time wasted and significant improvement in software quality.
Sandwhich a.k.a. Taylorism?
> the Manager is responsible for project success;
> the Expert decides how to do things;
> the Worker does the actual job; and
> feedback is unnecessary because the Expert has thought of it all already.
Except I suppose in your model the Manager gets the feedback?
https://entropicthoughts.com/handoff-waste-and-taylorism
Feels like it won't be long before software factories and software like e.g. Rinkata evolve enough to be able to do the verification part
How could they verify that the software solves the human's problem, and the human enjoys using the software?
To be honest, I doubt the current state of things will hold much longer. Current AI can automate virtually all of my work as a pre-AI senior and is rapidly closing in on being able to automate the types of work that have emerged post-AI too. I have a hard time imagining a world where a drop-in end-to-end replacement for a full software team is more than a year away. I'm sure I'll have a job for a year or two past that, but there'll be some small stock market bump eventually and a bloodbath will begin the second the C-suite takes a closer look at what their engineers are even doing anymore.
At which point the question will be: What are even the customers paying for? If they can get the product / service straight from their own chatbots, why pay a company for it? The companies that sell data, will probably somewhat safe - as long as they can have some moat between the users and data.
The death of SaaS could also be a thing in the (not so distant) future.
I’m planning out my 2030 career in idle moments these days. I think it’ll be something akin to a carpenter who builds custom furniture.
I think we’ve barely scratched the surface of what software can do if it’s completely custom built, specifically for individuals or small businesses who could’ve never afforded it, pre-ai. Home automation, health and disease management, all sorts of administration, conversational ai over your life’s data - most people have no clue what’s possible once it’s custom built. With close attention, good design, creativity, and without giving away privacy or independence.
In my limited spare time I’m building up this tapestry of custom services for myself and my loved ones, and having a great time doing it, while trying to learn how I can offer this as a service.
Judging by the reactions I get when showing people what I can do with my phone, and no that’s not available on Google play, I think the demand will be there.
Woodworking and making furniture is fun! However, having made both furniture (and ceramics) myself - I think it’s hard to make a living this way. The overwhelming majority of people are used to paying little for these kinds of things. In the case of woodworking a large piece of furniture can take months so you might expect to only produce 3-4 / year unless you’re very good.
Not to say you shouldn’t! But it’s worth being realistic about income compared to (current) software salaries.
Sure. But you're taking my metaphor farther than I intended. Most people understand the basics of furniture, and still prefer to pay Ikea rather than a carpenter. In the case of custom ("bespoke", ugh) software, we'll have to both define and communicate a new category of service. And the customers are likely to be the same sorts of people who will pay a personal trainer or get a private tutor for their kids - and I think that includes a lot more people than are willing to pay for custom furniture. But generally I do agree, this vision is far more hand-wavy than concrete at this point. It is, though, a good time for hand-waving right now.
Engineer as intellectual artisan, not factory worker
Ah I misunderstood your original comment then. I took it more literally as you eventually will become a carpenter (after software)
Fit doors and other joinery, you would be surprised at the amount of local facebook community posts I see, with people desperate for someone to re-seat a door or build a stud partition.
Professional joiners and carpenters are too busy with the big jobs to care about small 3-4 hours jobs.
That's my plan at least
> I think it’ll be something akin to a carpenter who builds custom furniture.
And who'd buy it?
>The death of SaaS could also be a thing in the (not so distant) future.
I suspect outsourcing is the first thing on the chopping block. Why bother with different timezones, cultures and languages when you can get your AI to do the same thing for less money.
Also the companies with leadership who bias towards outsourcing for cost cutting reasons, will 100% happily automate any Indian folk too.
Honestly I feel like a good analogy for AI in general is outsourcing. You pay a lot less, but you need to explain what you want very clearly, need to review the output, and need pass-fail success criteria. This is exactly the job of an outsourcing manager. At least there's no opposite-timezone hurdle and no language barrier.
if AI can kill SaaS, what limit the blast radius to only software. In theory any job that you can do on computer, AI should be able to do that as well.
And with the advance of robotic, what would we even do ?
> In theory any job that you can do on computer, AI should be able to do that as well.
It can, however, currently the issue is that sufficient intimate knowledge is required, and just a large complex code base and/or documentation is insufficient. However, intermediate representations to bridge the gap are on the way.
> intermediate representations to bridge the gap are on the way.
That's interesting. Are you working on these or are you talking about something else?
Yes, I am working on these. Particularly for representational systems - those that model external systems - rather than transformational software.
Robot repairman?
Consumers are unlikely to want to build their own software even if it's "easy" with AI. It doesn't really make sense to spend all that time vibe coding instead of paying $10/month for most people.
Businesses are more likely to build than buy, yes, but as we know from "Theory of the Firm" there are limits to this, it's not going to be correct to build _everything_.
So I mainly expect it to change what type of software gets sold, not the fact that it gets sold in the first place.
There's a lot of saas that costs a whole lot more than $10 a month and doesn't work exactly the way you'd want it to. Then it becomes very tempting to vibe code your own
> Then it becomes very tempting to vibe code your own
What about the Wall, though? The Wall (tm) is when you vibe code the software through prompts, and at some point, the accumulated cruft means that you no longer understand enough of the codebase/approaches to give useful feedback, so you delegate more to the LLM, then your costs keep increasing and eventually you hit The Wall (tm) where you can't make forward progress without breaking something else.
The models have definitely improved (i used to hit the wall in one session, now its generally weeks), but I'm currently working on a ridiculous tool replacement where even the creator has hit said wall.
Coding may be solved, but maintenance and maintainability (particularly where the software needs to be accurate) are definitely not.
We'll probably end up with the same number of software people doing a lot more.
Eventually, we'll end up in a world where custom software is mostly building harnesses and verification tools to ensure that the agents actually build the correct thing.
And we'll always need someone to interface between the users and the software, as most users don't really know what they want (c.f. faster horse etc), so the jobs will change but are vanishingly unlikely to go away (modulo the immanence of the machine god, but that seems unlikely).
Restaurants exist despite grocery stores selling ingredients. Thing is, I need to eat every day and food goes bad. Software that I need to use doesn't decay in the same way.
AI isn't a restaurant, it is StarTrek replicator, or frozen food heated on the microwave.
Star Trek Replicators need near-infinite energy sources to work, which we don't have. Frozen food still needs to come from somewhere. You still pay for Hot Pockets or Stoffers or Picards.
Yes, that is the gotcha c-suites keep forgetting about while wiping the work force.
most customers aren't paying for software at all
I think we're looking at the end of commodity software, the end of software companies, and birth of every single company writing their own everything. A proprietary soup everywhere, and 1-2 100X developers overseeing it all. And those few developers will spend their time in communications tutorials to get the staff to be able to explain what they want and need, so they can then make it happen for that company's exclusive use.
It just doesn't really make financial sense, a lot of software is not a significant cost to many companies. For example my company pays something like $50/month for newsletter sending software. We could easily vibe code an alternative but why bother? $50 is a rounding error in our business costs, even a single bug or outage in the replacement would make it a terrible decision.
On the other hand if you're a business with 50k employees and you have to pay $10/seat for something that's a much more compelling case to build in house (but then it always was). So maybe with vibe coding it drops the cutoff to a lower number?
The non-financial sense is that $50 newsletter still being a business, it becomes freeware / open source / cloned by someone that used to write that enterprise level, and is now struggling financially.
And that newsletter? How much of that is AI now, and do people read it?
We're in the midst of a huge cultural and economic shift, and where this ball of confusion goes is unknown, but it certainly will not stay the same.
Or commodity software companies that are small and experts and creating AI custom solutions extremely quickly
Why stop there? Any capitalist company will get rid of everyone else, and leave all office work to AI.
Where are they getting the money to pay for their AI, when everyone is without a job?
if everybody is out of jobs, then the labor costs will be so small that they compete with energy+compute costs. when they become lower then human labor would be reinstated. an economy consists of capital and labor. AI would transform this to only capital, producing then only for the capitalist. The system worked, because capitalists have right now a need for labor. Maybe there will be a dual economy, where the workers are producing their own products for consumptions without AI, while the capitalists are producing everything for themselves.
In such society only robots rule.
Accountability. Suppose you had the ability to have robotically built highway bridges. Who is going to sign off on the work? Who is going to be held accountable? The same with software. The people who don't like to write software today aren't going to suddenly develop a love for deep testing and validation.
And now we are controlling models that can be hacked in so many different ways and literally blow up companies. It used to be accountability the software wouldn’t break. Now it’s accountability that in creating the software we don’t ruin the business.
[dead]
> The death of SaaS could also be a thing in the (not so distant) future.
First off, this is wishful thinking.
Secondly, if it were true, good riddance!
I mean, the same reason they do now. Time, taste, and amortized costs.
Just because it's faster and cheaper and requires less skill, doesn't mean it's instant and free and requires no skill.
Whenever I see comments like this I wonder if we're even in the same industry.
Vibe coding does not change the fundamental economics of the classic software business at all.
I work as an analyst, and our group has - since last year - saved a couple of hundred thousand dollars worth of licenses for various analysis tools and packages, as we've leveraged LLMs (vibe coded) to clone those. We've always had the engineering skills to do it, but never the time or resources. Advancement in LLMs changed that.
And I suspect we're not the only ones doing this.
Who knows how powerful AI will be 5-10 years from now. I seriously suspect that in a decade (if not sooner) AI will be able to one-shot software that currently takes entire teams of engineers months to build, in mere minutes.
You are still thinking in terms of the past, if things go well we won't need to work anymore. Gotta think about how the world would function when only marginal amount of human labour is actually needed, not about how to get a job.
> if things go well
We didn't need AI for things to go well. We could all have a lot more if people were willing to collaborate and work constructively together instead of sucking rent from everything and tricking people.
Sure, I think that got overshadowed by the western capitalistic approach - it gave people a real possibility to work and get to a more comfortable place in society.
It seems obvious today that following it up to the extreme, optimizing technology and reducing production costs ultimately ends up removing humans from the process.
I wonder if the founders of capitalistic economic model considered this outcome, going to chat with ai about that lol.
"we don't need to work anymore" is not a good outcome for most of us
It means we become disposable to society
Not sure what it means "disposable to society" - there will be no longer an incentive to value someone based on how much they contribute in terms of wealth generation. I think people will still find their ways to do other things, after all we have time on this planet, gotta spend it somehow.
> Not sure what it means "disposable to society"
Look at how homeless people are currently treated for some ideas
> there will be no longer an incentive to value someone based on how much they contribute in terms of wealth generation
People need an incentive to value other people though. I don't know if you've noticed but "valuing other human lives" does not seem to come built in for us. History is built on humans killing other humans they didn't value, in order to enrich themselves
Until we reach true post scarcity I think most of us are in for a lot of hardship
> Look at how homeless people are currently treated for some ideas
I don't really buy this, at least not in developed countries - so many of homeless people don't even try to do anything, they chose this "lifestyle", they often ignore government and voluntaries programs society created to help them. And you can not enforce anything because we respect the free will, even when its completely enslaved by drugs and other things.
> I don't know if you've noticed but "valuing other human lives" does not seem to come built in for us.
I can actually argue that it does, even in nature among animals there are behaviours displaying kindness and value when it doesn't result in any profit. I think our (western) culture largely ignores these aspects.
> History is built on humans killing other humans they didn't value, in order to enrich themselves
Right, so maybe AI helps us overcome the enrich part, and as society we don't put that much value on being rich, this is where things become new and interesting.
> I don't really buy this, at least not in developed countries - so many of homeless people don't even try to do anything, they chose this "lifestyle", they often ignore government and voluntaries programs society created to help them
Even if they took advantage of literally every single program that is available to them, would they have the kind of lives you would want for yourself?
> Right, so maybe AI helps us overcome the enrich part, and as society we don't put that much value on being rich
Maybe one day, how many of us struggle and suffer and die before that happens? Do you think that transition would happen smoothly and without large social upheaval? I'm not really convinced
The dream of AI enabling a fully automated socialist utopia might happen one day in humanity's future but I have very little confidence it will happen in my lifetime. And I would really rather the rest of my life doesn't play out during massive social upheaval
> would they have the kind of lives you would want for yourself?
Right now while I can still earn money I prefer my conditions, because big changes didn't happen yet, eg I have to pay to get medical help, while in the future where AIs would efficiently replace doctors that would be one less reason, not needing a good location related to going to work etc etc.
> Maybe one day, how many of us struggle and suffer and die before that happens? Do you think that transition would happen smoothly and without large social upheaval?
These are complex questions for my level of understanding of society, I hope we can handle it well.
Let's look at the history of how ruling classes have behaved now shall we?
Hopefully this time the ruling class people are more educated, see the bigger picture, have everything to satisfy craziest desires and can see beyond selfish goals.
If they were capable of unselfish, bigger-picture behaviour they wouldn't be part of a ruling class. There is no such thing as benevolent power, every power will eventually be abused to the maximum practical extent it can be and every emergency will eventually become routine.
You cannot hope for a good absolutist king because good kings only exist in mythology. All you can do is design systems that structurally limit the power any one group or individual is allowed to amass over others in my view.
> every power will eventually be abused to the maximum practical extent
But what is the practical extent here? If you own unimaginable amount of wealth while most people are on lets say basic income, you can just afford anything, what do you get from abusing your power?
The same thing every cruel and capricious ruler has got from it, history is littered with example after example of the devil making work for unlimited wealth and power. That kind of wealth and power generally becomes self-justifying, humans have endless capacity for self-delusion and it is very easy for the powerful to convince themselves that their selfish actions no matter how inexcusable are actually generous and aligned with the public good. When it comes to outright oligarchy, wealth and power are simply aspects of the same underlying phenomenon.
At any rate, your common masses on basic income are still living under illiberal conditions even if their shackles have been thoroughly padded. Liberty isn't merely the freedom from interference as neoliberal types like to claim, since ancient times it has been understood that liberty is the freedom from domination. According to the classical formulation a renter in a liberal democracy is still not at liberty because their landlord may turf them out on a whim, the wage labourer is not at liberty since they may be laid off at the whim of a board who neither know nor care to know their name, anyone whose living circumstances depend on the continued whim of even a benevolent master is not a free individual. Under this framing, the people on basic income would be profoundly unfree since their entire lot in life depends entirely on the benevolence of the oligarchs.
There is a good reason that even the Romans understood this danger even as they failed to prevent it. Republics should not suffer private princes, it is a lesson we have been failing to learn for over two millennia.
I don't think elites are ever benevolent, but sometimes they do limit their abuses out of long-term self-interest.
A good example of that is the Roman Republic. The elites created a system that gave land out to retiring soldiers in parcels that were large enough to make them middle class farmers but not genuinely wealthy. That let them churn out absurdly large amounts of heavy infantry that no one else could match. That made them slightly less wealthy in the short term than if they had just taken all the land for their own estates, but it made them immensely wealthy and powerful in the long-term.
They were also very welcoming to outsiders compared to their contemporaries. Conquered territory was basically fully Roman within a few generations (though still keeping many local customs).
Contrast that to Sparta where the elites held onto every ounce of power with an iron fist, keeping most people in slavery, and made it virtually impossible for anyone born outside the elite to join it. They ended withering away to practically nothing, sometimes fielding armies with only a handful of full citizens commanding it.
The early 20th century US is another good example. Fear of communism led to many elites supporting the expansion of worker's rights and the institution of a social safety net (FDR and JFK were both from very wealthy families).
Lol
C-suite is as likely to be automated as engineers. It will probably be AI firing engineers on behalf of shareholders.
No. The C-Suite is safe, very safe. I participated in an AI survey of what is the information that a C-Suite operates upon, and it is largely unwritten information, it is a suspicion and manipulation game with all the other c-suite players, and the middle managers beneath them as the battle ground. There are Peter Principled members that maintain their positions by executing the corporations legally grey dirty work, and then there are serious 100x non-engineers that can launch $100M initiatives in short notice, with successful experience doing so to create divisions and more.
I know the software engineering world like to think the C-Suite are disposable, but that is only because you really have no idea what they really do.
i think the point is that engineers can go build products with small teams and no need for a c-suit. not that large corpos would replace the c-suit with ai.
Do shareholders know what they do? The C-Suite has bosses too.
Middle management should be nervous. Plenty of people whose job it is to ask "is it done yet? when can I have it?!" -- they could be replaced today with better tooling and processes.
> C-suite is as likely to be automated as engineers.
Nah. C-suite has (a) money and (b) legal and fiscal obligation. You literally need a human being certain legal operations, like incorporation, salary, and lawsuits.
I think a lot of people thought that every step of the way with the progress of AI or automation in general. Unless that C-suite is all shareholders with majority or near majority control of a company, I would think the passive shareholders (in a public or private company) would want the savings in cutting all of those high salaries to zero for a likely superior outcome with AI.
I wouldn't count on lawyers being deemed necessary by the law for as many actions as they have been so far.
> would want the savings in cutting all of those high salaries to zero for a likely superior outcome with AI.
Maybe. Or maybe the C-suite salary will be cut to $50K/yr. That would make them marginally more expensive than AI, but for that price the shareholders get an extra layer of insulation should things go awry (and they won't be held personally responsible for corporate malfeasance).
> I wouldn't count on lawyers being deemed necessary by the law for as many actions as they have been so far.
Maybe, but that's pretty far down the road and would require restructuring the whole jurisprudence system.
The first truly independent agents will operate a cybercrime syndicate.
You need *a* human. And that human will be delegated to clicking approve on AI decisions.
Who is C-suite going to blame for fuck ups and failures if the company is just a suit and a laptop?
Oh my poor, naive programmer. You don't know how power and status work, right?
Let's just say that as a shareholder I'll support keeping headcount and expenses minimal.
People have been saying this for years. They’ve also been saying “have you tried n” model yet as well as “wait till n+1” model is here as the author mentions.
I just had fable 5.2 (high effort) review a change. It hallucinated 2 problems that I was only able to correct because I also understand the code.
LLMs can do any job _in theory_. It’s not a very interesting argument. Their effectiveness at various jobs is much more interesting.
You're holding it wrong. You saw the Fable 5.2 marketing. You're supposed to use it in a swarm of 1000 Fable 5.2 agents. Much better results. The RLCR training gains are real. You do have your own datacentre right?
> I also understand the code.
and the product, the team, and (no matter how shallowly) the organizational pressures and politics. Whether we like it or not, those are part of the problems developers solve. Even something as rudimentary as knowing what is feasible to test given your deadline isn't something you can encode in a model[1]. Code does not exist in a vacuum.
I say this even with the latest models, it seems as soon as your development practice falls out of the ideals written up in blogposts of the previous ZIRP decade, you have to course correct. I use this term because the AI is technically correct (the best kind of correct) but practically useless, if not also irresponsible.
Recent anecdote (literally today): we've been having Claude write exhaustive test suites to eat at the coverage backlog. Lately, we noticed that tests in develop are even more flakier than usual. Our whole suite runs in 30-40min but lately it could exceed an hour and then fail.
It seems Claude took "exhaustive tests" to mean "exhaust the compute resources of the Jenkins node". What's worse is it wrote those tests for features that have been disabled and would soon be rewritten anyway.[1] I've always said that LLMs are great for generating test suites and this is the first time an LLM "failed" this assignment for me.
Code is cheap. The right code (or lack thereof) for the right situation takes meatspacebrainpower.
[1] The solution here is to run your company/team 100% as agents so they have the full context of project timelines, resources, customer feedback, etc. ala stevey and gtan but my reasons why I don't believe in that has already been articulated elsewhere.
> I've always said that LLMs are great for generating test suites
I always found that LLMs aren't great at writing tests, they have no notion of CI maintainability and they write a lot of "is 1 = 1?" worthless type of tests.
I've noticed they also love to write jest/vitest exhaustive input cases when one will do the job.
> To be honest, I doubt the current state of things will hold much longer. Current AI can automate virtually all of my work as a pre-AI senior and is rapidly closing in on being able to automate the types of work that have emerged post-AI too.
reading HN these days feels like constant gaslighting. this is so far from my reality that i wonder what folks who are saying this are working on
Every time I see one of those comments, I think about two open source libraries that we use at [DAYJOB]. Both recently released a "new version" that is actually a full, ground up rewrite of the entire project.
In both cases, they have:
1. Weird bugs that a human would never write (e.g. doing async tasks in synchronous functions).
2. A migration guide that is so verbose as to be worthless. Think 50+ pages of text that give every minute change equal weight.
The old versions had some warts, but they worked fine. The new versions have been inflicted on us, and it feels like the intellectual version of digging a hole to fill it in. There was absolutely no reason for it.
I'm sure that the people in charge of those projects believe that they're massively more productive, and that what they're doing is good. But it's worse than useless, it's just a waste of time for everyone involved.
So whenever I see someone claim something like "I used to be a 10x developer, now I'm a 100x developer," I have to wonder how much they're just inflicting garbage on everyone else and they just don't have the self awareness to realize. The type of person to show up at a job, crap all over the codebase, and then leave before it catches up with them.
AI is the ultimate generalist. You can make it do a lot of things well by asking the right questions. In a generalist job, many people can ask those questions.
I don't use AI that much. When I do, it's often surprisingly competent in the topics I'm a specialist in, and it can do a lot of good work quickly. But other people in other organizations still continue building upon my work instead of asking AI directly. Apparently because they don't know what to ask.
Nobody knows how long this will last. Hopefully long enough that I can afford to retire, because I'm old enough that I would face ageism if I had to find a different job.
And if we get there, we will probably have a serious political debate on how many digits should there be in the wealth tax rate.
[dead]
I think people who were in mostly bs kind of roles, honestly.
I have become the “AI” guy at my place, unfortunately.
For some reason I can “hold it right” and others cannot and for the life of me I cannot figure out what it is. I am not more intelligent and certainly not an amazing programmer. I currently suspect I am better at separating what is and is not important. I find many people miss the forest for the trees and overindex on irrelevant details. The more technical they are the stronger this tendency seems to be.
I regularly, like almost daily, encounter colleagues claiming it cannot do something I either did the day before or will just demonstrate right in front of them. The response is usually rationalizing and claiming it wasn’t “that hard to begin with” AKA shifting goalposts.
I have never once encountered a work problem which I truly found impossible to automate, which is insane, but the institutional inertia is not worth overcoming. Devs are a thorny bunch and usually think they are special. I am a recovering one.
I think many are in for a rude awakening.
> For some reason I can “hold it right” and others cannot and for the life of me I cannot figure out what it is.
I don't think we talk about this as much as we should. I've been interviewing devs as part of [DAYJOB] and there's a lot of people who cannot get anything useful out of Claude. Forget features, we're just talking bugfixes.
I also don't know what's different about the behavior. My best guess is people are overprompting, adding a bunch of noise that interferes with what they're trying to achieve.
I don't see a rude awakening for a technically competent person not having adapted to this year's technology yet. It's pretty weird to expect a functioning development process to suddenly stop working or malfunction just because an even cheaper or faster development process could hypothetically be adopted today.
> suddenly stop working
I think we might already be there with security. Exploits, hack attempts are up and likely continue to stay high. Traditional development process could work but it’s probably more engineers than management wants to spend on a cost center.
I roll my eyes every time I read these comments as well. Can't imagine saying something like this without being an absolute moron with an incredibly underdeveloped model of how the world, and business specifically, works.
The worst part is the morons upvoting it as well.
> this is so far from my reality that i wonder what folks who are saying this are working on
HN these days is mostly literal bots, AI company shills or people working on some shovelware AI slopware themselves, so they constantly throw out these generic statement as if it's fact. I've been reading this same crap here for the last 3 years, how "Wow, now with Newest Model Mk. 3 (tm) from Anthropic (R), everything is automated!" with very little variation.
It's getting pretty tiring listening to the shilling to be honest. The best part is that when you actually look into any of this crap, or read any of the comments [0] from people in the industry you realize how fucking amateurish and ignorant they all are
[0] - https://news.ycombinator.com/item?id=50408126
I'm guessing it's the copy-paste CRUD apps that used to be plagiarized from stackoverflow. I'm with you, I've never seen an AI agent do good work.
Idk, I have my fair-share of copy-paste CRUD work and I have yet to see good work come out the other end outside of a few very very isolated and constrained examples. And before "you're holding it wrong", I am constantly having to firefight and rework code produced by the most AI-pilled people at work.
[dead]
Skill issue. It’s just that simple.
Nothing devs do is interesting or hard. “Rocket trajectory optimization” is not dev, it’s math. Dev work is very boring and very simple.
I regularly deal with people that think they have amazingly hard jobs and it usually boils down to it being undocumented, arcane and/or math-y. The software itself is always uninteresting (as it should be).
Everything is “CRUD”. Name something that does not amount to taking input, performing a transformation and producing output. I am so tired of this overused thought-stopping cliche.
> is not dev, it's math
As if math has no interesting problems to solve, and training data exists for every possible scenario.
> Everything is “CRUD”. Name something that does not amount to taking input, performing a transformation and producing output.
This is true in the "Reductio ad absurdum" sense, but software has to be written to drive hardware that interacts with the real world. Physics is arcane and/or math-y, and many of the problems that are still outstanding in the field of software engineering (or engineering as a whole) are a result of this. The world is full of unknowns, and inferring how those work with a pattern-matcher will only get you so far.
Perhaps other people at other companies are having different experiences than you, rather than "gaslighting"?
You know, Elon and the "superhero" DOGE team said they "couldnt imagine" why most of Govt was required. What happened?
[flagged]
A lot of people died.
[flagged]
Guess because they're not white and live in your zip code it doesn't count.
Security will always be a risk with AI tooling. The only way to avoid that risk is intelligent human reasoning. No, not human-in-a-loop. It’s human defining boundaries, carefully, and allowing agents to operate within them. Failure to do so will always lead to disaster given enough time.
Then who directs the swarm to create the most value? An executive can’t. They have other things to do so they delegate. That person will be an engineer and their attention is finite too so they will delegate. To other engineers.
Here’s the best part. You write back and say I’m wrong. Guess what? That’s engineering too. Some people have to sort out all this nonsense and the stakes are higher than they’ve ever been and now engineers are supposedly not relevant.
But yeah when you get everything under control, what’s the day to day look like? Pretty freaking cushy. And then it relies on intelligence to figure out what next. Clients always want things done and having those conversations won’t get much easier with AI. Keeping up to date. Jesus I’ve exhausted myself writing all this. So much work!!
And your reaction is to accelerate the process by saying exactly what BigAI wants to hear? Totally believable.
What happened to all that "RTO is to keep to C-suite happy" talk around here? How are they going to prop up real estate or feed their egos or whatever the conspiracy was.
That's where my 1-2 year buffer comes from.
[dead]
Something I was thinking to myself last week – "if I believe SWE is more or less dead then why am I so busy?"
Clearly there is still some work to do, but the work I do now is very different, and for me, much less enjoyable.
I'm not solving challenging logic or algorithmic problems anymore, but basically just iterating on features with agents or getting them to look into things for me.
It sucks because I used to be very good at these things, but my edge against less competent SWEs doesn't really exist anymore.
More annoyingly, people I work with are producing a lot of code that I think the design of kinda sucks, but agents can force the design to work and so it works. I suggest we do things another way, but I'm now just slowing things down and it's hard to argue against the comment, "if that becomes a problem we can just get Claude to refactor later".
I guess what I'm saying is that I think think there's a really crappy job that will still exist for a while around managing and contributing to these agent-generated codebases. But skill is less about your technical abilities, but how good you are at hooking up agents to things and getting them to solve problems. This doesn't interest me at all though. My desire to spend a lot of time design and thinking about problems, and deeply understanding the codebase I'm maintaining now just work against me.
I think there will be SWEs by some definition, but the job as it existed a few years ago is gone. Not even saying that's a bad thing, but for someone like myself, it may as well be dead.
The move now is toward "software factories" where sucky design is no one's responsibility.
Eventually the whole thing will grind to a halt or fail catastrophically (that's life), but it will seem very good while it's working.
I don't know if there's a real role for engineers beyond designing and tuning the factory itself.
Of course, such a thing is only tractable on highly factorable platforms. Well designed incumbents have an edge for the next couple of years or so, until others catch up.
It gives me no pleasure to say this, but I think it will all be automated in the end. Engineers will be more like rare consultants who inspect and validate very high level solutions.
What does catastrophic failure look like for the software factory model? I struggle to imagine substantial risk the pattern introduces that wasn't already a risk in traditional engineering teams.
If anything, I've seen the devs chasing software factories be far more intentional about codifying their architecture and design principles. If they don't, it never gets off the ground.
we might be in the java applet era for all we know.
it’s a mix of (at the time) cool, shit, and oh my job!
who knows. and i’m sure someone will point to some other tech.
i only picked java applet because all the vibe slop looks the same. and you know instantly what you’re dealing with lol
Hey! I’m the one who wrote it. Saw you asked for a non-x link -https://shahar.blog/2026/11/08/on-being-an-engineer-in-2026/
https://nitter.cf/shaharglazner/article/2119796029256032278
A readable link for those of us without Twitter/X account
Thanks - we'll put that in the toptext as well.
https://shahar.blog/2026/11/08/on-being-an-engineer-in-2026/
Much better. Swapped above. Thanks!
Also, I work as a data analyst, with previous experience as a software engineer. In our office, traditionally maybe 10% of our analysts have enough real software engineering experience to develop any "serious" software products. The rest are domain experts that at most use programming for developing methods and calculations/simulations.
In the past 6 months, we've seen non-engineering analysts leverage LLMs (Claude, Codex) enough to launch actual products. They've gone from 0x engineers to at minimum 1x engineers.
Prior to this, they'd develop methods, scripts, etc. in Jupyter notebooks, R, Matlab, or whatever they'd be comfortable with. And if it turned out to be useful, they'd work with engineering to develop those into products. Usually a small team of software engineers would do that.
But, alas, now they're able to almost completely bypass the engineering teams.
> But, alas, now they're able to almost completely bypass the engineering teams.
We're wrangling this beast currently. I'm on a very small team, and it's a difficult new frontier. We're maintaining the "core" software and giving them tools to build safely with their new fancy tooling.
It's not going perfectly, but I suspect this is the future. Everyone is a developer to some degree now, the backend/core systems needs to be plug and play.
What's really difficult is maintaining focus (when new 'production' applications show up over the weekend), knowing when to stop building a POC, and keeping an understanding of the whole system as it just scaled up significantly into some newfangled satellite based architecture that we're trying to standardize...
I love watching people be able to build tech, as it's previously been gated on technical expertise. However, it came overnight and my team was NOT ready for the speed...
Then you work with more capable data analysts than I did. When I was a DA, my colleagues knew dbt and SQL. They didn’t know Python
If you can’t make things right when your agent goes wrong, then yeah you’re not adding value. If you can’t tell the difference between the agent hallucinations and the useful stuff, then yeah you’re not adding value. If you can’t steer the agent in a useful direction instead of just standing back, then yeah…
You sound dismissive, but taking your description to its conclusion, the engineer value can be approximated as:
For many engineering jobs, P(AI mistake) is very low and decreasing, and cost of AI mistakes is also usually (consider the cost of a bug for an average engineer; unless you’re building avionics or working on MRIs or such, usually very low).So even if you can “make things right” or “tell the difference” as you say, the value of any given engineer is dropping.
> For many engineering jobs, P(AI mistake) is very low and decreasing
This is not my experience. Well, at least not for what I'd consider very low.
For example, I have a standard task of adding support for a well-specified regex scheme as a dependency-free module to the Wren language (which grew out of actually needing it at one point). When done autonomously, errors still occur, some of them very basic. For example, trying Fable 5.2 at medium the other day produced the following:
- memory unsafe behavior on 32-bit architectures (the good old problem of size overflowing on an allocation call).
- O(n^2) behavior for searchAll() and replaceAll() operations.
- Various minor subtle semantic issues, though none of them particularly critical.
This is a well understood problem, and in general, Fable's construction (Thompson NFA by way of a Pike VM) was in principle correct, but it still had these pretty basic problems. (Other frontier models produced their own, e.g. a use-after-free bug or a combinatorial explosion in the regex compiler for some types of regexes.)
(As an aside, so far I don't consider Fable 5.2 an improvement over Opus 5.5 in coding ability, more of a sidegrade; while it is smarter and can do some hard, but also pretty esoteric stuff that Opus fails at or takes much longer with, for regular coding problems it's neither better nor worse, more ... different. And sometimes Fable seems to outsmart itself, producing a seemingly clever solution from its vast knowledge that is actually worse than the standard approach in practice.)
One area where I absolutely do not trust even frontier models yet is parallel and distributed programming. I absolutely value them as a second set of virtual eyes for tricky situations in this area, but I do not trust them to get things right, and even if the solution is correct, it's often inefficient or convoluted.
Your work sounds interesting, but I hope you agree that it’s not representative of the median engineering task.
Futhermore, what would be the organizational cost if these bugs were in the codebase?
> Your work sounds interesting, but I hope you agree that it’s not representative of the median engineering task.
This was actually from a personal project, and my point was not about the type of task, but the types of errors I am still seeing from modern LLMs.
> Futhermore, what would be the organizational cost if these bugs were in the codebase?
The particular errors that I highlighted above are what I call "landmines". They may not manifest at all depending on what you do, but if they do, they can hurt a lot, especially in production.
Most devs don't even know Wren exists, and their job can be easily automated by Workato, Boomi, SitecoreAI, Optimizely Opal, PowerPlatform, ....
That this is in the context of Wren is completely inconsequential here. It was just what motivated the original task (because of lacking functionality). What I was pointing out was that fundamental coding errors such as memory safety errors or O(n^2) code still happened with what is considered one of the most advanced models we have, even though this was overall pretty clean code.
In that case, still quite similar to the quality of offshoring projects with Human Intelligence, thus management hardly cares.
No disagreement (except that I wouldn't generalize it for all management), but I was only arguing a limited point about the likelihood of AI mistakes.
It really depends if you consider "AI mistake" to be "code logic error" or "code doesn't match real world expectations". The former is getting quite rare I agree, but the latter is still extremely common.
> For many engineering jobs, P(AI mistake) is very low
Not remotely the case. LLMs make mistakes very frequently, much more frequently than humans do.
source? I'm wondering at the curve of those mistakes actually, I doubt modern LLMs make mistakes at the level of the worst engineers I've worked with.
My last project had 2 guys from India and one guy from a consulting company that I take any LLM over them. The guy from the consulting company spent more than a month trying to implement a design that actually was not implementable. I obviously also blame the supposed professional design company that produced the design, and I mean it took me about 6 hours spread over a day and a half to figure out waitaminute this is actually not doable! But I have to wonder how badly would an LLM have done. I sort of think the same, because probably an LLM is never going to announce - hey, this design as stated is impossible, some parameter needs to be changed.
What's the doubt based on if even in your only example you only wonder, yet still conclude that LLM would not do better??
my doubt is based on the fact that I do not have data to draw a conclusion one way or another, I have what is called a supposition, and my supposition may be affected by all sorts of fancies and ideas that prey on my mind but are not established as true.
I had no doubt that your doubt could be based on nothing and affected by anything, that wasn't the question.
it's worse than that, as you need to also compare
human mistake, AI spots the mistake
Human mistake, human spots mistake
AI mistake, AI spots the mistake.
Everybody is always talking as if they don't make mistakes, but every software engineer does; it used to be that human caught those mistakes and fixed them, but if AI catches those mistakes and fixes them, then that is also a decrease in value of human.
Human - Human and AI - AI mistake and spotting thereof basically cancel each other out, or can be considered no mistake. Human - AI just speaks in favor of AI, and my comment was responding to somebody dismissive of how crushed engineers can feel by AI.
> This naturally leads us to think that this is it. We, engineers, are doomed.
I find it _very_ unpleasant that all these people constantly talk about a "we" that does not exist.
That person is worried, and/or they have seen other people in the industry that are worried. But they've wrongly extrapolated that everyone in the industry would feel that way.
It harms the writing. It shows that the person is not adequately grounded.
Or, tbf, maybe, made mental shortcuts to reach people. But they've sacrificed truth to connect there, which is bad.
As the one who wrote it, I thought about using "we" vs "I" (and in other places I did use "I"), but its bit overthinking about what I tried to say. But anyway will be take more caution next time
[dead]
What demotivates me the most is how easy it is to copy stuff these days.
Suppose we still need judgement and taste to guide AI to a good result. E.g., even if the AI has these itself, if the person guiding the AI has bad judgment and taste, the AI will comply, and build the bad thing because it tries to be aligned with the user.
But now, someone else’s _good_ judgement and taste is worth only so much because it takes an afternoon and some tokens to just “do it like they did it”.
I share this sentiment. But OTOH, if something is so easy to copy maybe we just have to accept that it's not that special anymore. Isn't excellence in a sense *defined* by ability to consistently achieve results above average? So the only consequence of the average shifting upward is the threshold for excellence shifting accordingly?
What constituted excellence a year ago is trivial and common today, so maybe excellence just shifted somewhere else, and the realization many of us are having is that our valuable skills are now just average, and the top has shifted upward?
> Isn't excellence in a sense defined by ability to consistently achieve results above average?
I agree. If you're not consistently outrunning the competition, or convincing users that the quality of your work will always be slightly better than the copycats, than maybe your skillset is just not worth what you think it's worth.
This is essentially what I’m hoping for, that the whole industry will become more ambitious. But it feels like it will take some time for a new leadership to emerge. For now the mantra seems to be “build the same product faster and cheaper”.
Shifted upward to what, though? What is still hard to someone with enough tokens to burn?
Maybe hardware? Agents workflow design? Faster/more accurate models? Using our preferences to decide what to build and our agency to actually do it (poetically speaking, being the gradient in the agent optimisation process) ?
I don't know! But maybe that's how it feels like to be far from excellence again; maybe that's how it feels like to be out of our depth again, just like we were at the beginning of our career?
I don't want to risk stretching the analogy too far, and it might be that agents make all cognitive work obsolete. But surely there'll be something that defines a capability frontier? The alternative position is to believe that there just won't be a frontier anymore (I guess that's the RSI/AGI/ASI position :shrug)
Not OC you replied to.
I had similar question as you and conclusion as OC.
My answer maybe abstract in some ways, but:
- something that requires network effects, e.g social media, even twitter with megahitler stuff is going strong. Even though twitter's base app model could be done in 30min by LLM;
- something that maps to infrastructure in someway. Good luck copying Cloudflare, residential proxy provider, established email provider or VPS with LLM in a weekend. Though parts of it are definitely easier;
- something that is a platform (adjacent to first point). Google's search platform, ads, docs, youtube, music, email, business tools, etc. Similar to Meta (seemingly half of the world's small businesses control their business with whatsapp) and others. Though parts of those platforms can be disrupted with LLMs;
Right, but all of those examples are - if you’re already big and established, that’s your moat.
Yeah agree with this, these seem like genuine moats for now (we'll see whether this is still true in 6-12 months)
100%. Any ideas we have to spin up a new artwork game, analysis, or software - how long until someone says: copy that and change it 1% and now it's mine.
To those who sing the lullaby "AI has bad taste": You can tell it what to make. Your taste or any hypothetical taste can modulate everything it makes. There's no taste-moat here, folks.
Right, so, although taste is an advantage, the time you have to exploit your first mover advantage has decreased dramatically to almost nothing.
At first I thought tacit knowledge would be a differentiator: LLMs are trained on writing, so they can’t learn things that have never been written down. But tacit knowledge is exactly what reinforcement learning post-training is about. So (just as with humans) the limitation is in providing a rich enough environment, with good enough feedback, to train in the tacit knowledge.
The trick is to remain intellectually stimulated and challenged.
As someone who worked as both an engineer and a product manager before AI, being an engineer increasingly feels like being a product manager, just with a team of agents instead of a team of humans. Still some aspects of "tech lead / engineering manager" thrown in to understand the technical details for when the AI goes wrong, but less and less. Next we'll see how much of the product judgment will get automated too...
AI is a boon to engineers, and a coffin nail to programmers
> It started with giving up on writing code. Then it expanded to pretty much any writing - Slack messages, documents, emails.
I find it very strange that observability is always left out of these types of articles. Does nobody care about what happens in production and how customers are actually using what you develop?
Brave of you to assume that the slopsters give a shit about any kind of long-term maintainability or sanity in the products they're crapping out
Production? Don’t we do our testing there?
I'm curious about whether LLMs could continually learn from the feedback they receive while performing some tasks without translating those experiences into text. But at the same time, I'm reluctant as tacit knowledge feels like one of the last moats we have left.
The OP tries to answer the question "What makes some engineers above average?" and ends up with interesting concepts, but I think they're missing a key part of the question that already has a scientific answer.
Tacit Knowledge, intuition that comes from experience, times when something seems off but you can't quite put your finger on why - science has studied this! Specifically cognitive psychology. The answer, in modern terms that apply to both human and artificial intelligences, is training data. A good summary of this is "The Expert Mind" (Scientific American: https://www.scientificamerican.com/article/the-expert-mind/, PDF and HN discussion from 2006: https://news.ycombinator.com/item?id=41282574 )
tl;dr: experts have lots and lots of practice.
(The research does go into a lot more detail, for example time-on-task is not the same thing as practice.)
This has implications for the CS degree of the future, because it seems a lot of colleges are "pivoting" to AI, for example their Java module might as well be renamed "Using AI to write Java". To me, this sounds like exactly the kind of thing that is _not_ practice. It's the difference between trying to do maths problems by looking at memorising the solutions without (trying to) solve the problems yourself first. I'd really like to try the opposite way: a CS degree where you get deep into the fundamentals and are forced to write code in a sterile AI-free lab environment. My hypothesis is that this will produce (e.g. Java) programmers who will pick up AI-Java-coding quickly enough, but also have the intuition to step in when the AI code seems "off".
I think the author has confused "Software Developer" with "Engineer." The two are related, but distinct.
It's all about trust
I don't think engineers job is done, but I do think (and feel) that we are being pushed to the edge. I work constantly on multiple stuff. I constantly find my self being shifted around from task to task because "this one is more important" and "that one is urgent".
My point is if they'd just let us work properly, we might gain the advantages they all think we do, but right now, its shipping on the cost of quality.
I mean a running joke in the office is "whens the last time you even run the code?"
AI is NOT coming for your job. Not your whole job...
It's coming to take away the roles in your job that can be done by AI.
It happened before and it'll happen again. Humans adapt - there's more work now than ever. Some examples from the 20th century...
Big companies used to have printing departments to which you'd send a print job, walk over and pick it up, nicely stapled. Nowadays the office printer sits there 24/7 awaiting your print job. You have to staple it yourself though.
Big companies used to have secretaries & clerical staff who'd type up your letter and send it off for you. Now we have keyboards and email. You have to send the email yourself though.
Big companies used to have "code typists" who typed out huge amounts of code. Now they have agent managers who delegate the typing. Getting good output is a challenge, but asking for the output you need is so much better than RSI from typing and so much more effective than debugging based only on one's puny human knowledge, that we're suffering through. You have to assure the quality of the work yourself though.
It's strange seeing so many people take to their blogs and loudly declare themselves unemployable. I doubt anyone'd hire this author and that'll make their prophecy come true - for them.
A cautionary tale indeed.
I wonder what the impact of LLM coding is on software outsourcing. A lot of software jobs were being outsourced overseas for cost (supposedly). LLM coding feels like a tactical nuke dropped on that entire model. I haven't bothered to look it up but I would guess that the entire outsourcing model is about to disappear.
You forgot to mention what happened to those in the printing department, secretaries,....
They got new jobs?
The lucky ones
Every single one remained jobless forever and ever and never retrained or tried a different job or got a new career they just got really poor and died in a gutter.
Sheesh.
Actually that is exactly what happened to plenty of them.
You folks don't really have the notion what is living in regions where there are not that many job opportunities, or social support, right?
> plenty of them
Citation needed
Life experience
I don't buy the reductio ad absurdum, which hinges on the assertion that some developers are still 10x more productive than others due to differences in their engineering potential. I think that A) however you're measuring productivity is probably not capturing the full story, and B) one of the most important reasons people have differences in productivity is because not everyone puts the same amount of effort in. It's not necessarily because they don't have the ability to stay off of reddit and work longer hours and prompt more furiously, it's because not everyone is equally motivated to do so (hence the entire field of management).
I also don't buy the idea that there is special human stardust that gives us permanent irreplaceable economic value. Even if the tacit knowledge that the current generation of engineers were able to build up in the years prior to ~2025 does give us value that won't be completely replaced for a long time, why would you assume that the delta between what you can do and what claude can do will still be worth $200K in 5 years?
What if the market for human consumption becomes human made in the next 5 years? I could replace grand oak furniture with IKEA night stands worldwide until the world regrew its desire for quality
> If anything, 10x engineers have become 100x engineers. And the average engineer is still average.
100x more output at 0.5x quality.
I'm curious how actual engineering fields are affected - this highlights the imprecision we use when we call programmers/developers "software engineers."
Other engineering disciplines, presumably, don't change so drastically from one year to the next...
No because the cost of fucking up is usually that lives are lost.
But the cost of fucking up is really high in software, too. Countless people have had their lives ruined by hackers exposing their personal information, yet security is almost always an afterthought.
And there are also cases like Therac-25 in which people have actually died, but I'd like to think that medical software is more regulated now.
Other engineering fields usually have a much longer feedback loop. Software can be verified and tested relatively easy.
On Being an [Software] Engineer (i.e. AI agent manager) in 2026
Am I the only one who finds it weird that someone whose profile says they are a CTO starts their article with how AI meant that they had to delegate away debugging and writing code.
Weren't they supposed to do that anyway?
I need to update my profile, I’m Tech Lead at Elastic after we got acquired
Oh he means programmers. I thought he meant real engineers.
I think that distinction is dead and buried. Even your most serious “software engineer” has little in common with most engineering fields.
No, the distinction is more important now than ever. We need an engineering discipline behind the construction of software if we are to have any hope of squelching the tide of crap that resulted from LLMs being allowed to generate code.
Rigorous testing of materials is part of the process that allows structures that last human lifetimes to consistently be constructed, on time and on budget.
Programmers with title inflation are a dead end, their time is over.
The problem is that we don't have software engineers and we badly need them. The blog post is talking about a phantom, something that never existed, fading away.
Moreso now with LLMs sadly.
Agreed, and as a proper engineer, I can also assert we only need a few per construction site, and now that the workers can be automated away, what do the workers, and the surplus of engineers without a construction site?
Replace "construction site", by whatever relevant engineering work.
"Maybe one day agents will have their own experiences, too. Maybe they'll develop intuitions they can't quite explain, just like we do."
Imagine believing you have stumbled on to a benign consciousness and enslaving it to write shit code.
"You see a turtle on its back, but you're not helping"
We’re awash in these claims without evidence. A so-called “founder” of a company with no reputation or economic impact to speak of claims the entire world has changed and they’re 100x engineers now, and the code has never been better and the features have never shipped faster. But there’s no evidence for that claim in the economic data, no explosion of products, no new features in any product I use that isn’t just a shitty chatbot integration. Where is the 100x more software going? Why does the internet appear to suck more than ever? What are other people seeing that I’m not? I see 100x more “thought leaders,” 100x more text to sift through, 100x more lame resume-padding repos and budgeting apps, 10x more grift and vitriol online, 10x less collaboration between developers, but on the other end of that extrusion, people are just lonelier, more tired, more stressed, and pissed off about how shitty software has become.
> What are other people seeing that I’m not? I see 100x more “thought leaders,” 100x more text to sift through, 100x more lame resume-padding repos and budgeting apps, 10x more grift and vitriol online, 10x less collaboration between developers, but on the other end of that extrusion, people are just lonelier, more tired, more stressed, and pissed off about how shitty software has become.
The goal unto itself. Chaos, supported by a socially-reinforced shell game.
why so mad though? just my thoughts about smth I was thinking about. feels like u take it too personal
[dead]
Have no fundamental drive, no fundamental agency (ironic for "agents"). Whereas we have a will. That makes all the difference
AI has breadth, humans have depth.
Many of you were never engineers in the first place. It may hurt, but that's the reality.
And no, orchestrating agents to do your bidding doesn't make you one either.
No one who works with software is an “engineer”. An engineer is someone who is at the intersection point of math and physics. If all you do is designing OOP classes or React modules, then you’re not an engineer. These people must find a different term because it’s just completely inaccurate here.
Slop shaping is our destiny
[flagged]
> Maybe one day agents will have their own experiences, too. Maybe they'll develop intuitions they can't quite explain, just like we do.
Machine / program / software cannot have these things in my opinion because this stuff is produced by human consciousness which we cannot even fully explain yet. Someone has "to teach" the agent to have this stuff, but we don't have that knowledge. And also "intuition" is a feeling (like love, hate, sadness, joy) so no agent can have that until they have the limbic system, prefrontal cortex and other stuff.
As far as coding goes, I often had to make architecture decisions that were just gut-based on decisions that appeared to be a coin-flip at the time, an intuitive awareness of use cases that I wasn't able to articulate. I figured that was just some deeper pattern matching on a sub/un/super-conscious level.
Not everyone knows how the brain works and many believe it to be one organ.
LLMs learned from their training data, and inferred plenty of things that were implicit in that data. We don’t have to spell everything out for them - they learn, much as we learn.
Intuition is essentially a form of system 1 thinking, which is an LLM’s default mode of operation, before it gets into chain-of-thought loops and so on. You input a prompt, it’s processed through an enormous network of weights, and a result pops out. A model can’t tell you how it did that.
That’s a big reason why research has shown that LLM chain of thought doesn’t necessarily reflect its actual process for arriving at conclusions. It’s exactly “intuitions that they can’t quite explain.”
it doesn't "learn" it's a token prediction algorithm.
There’s no functional difference. You’re confusing the implementation with the results.
For all you know, your own brain works much the same way, at a high level.
"We can't explain it therefore we can't create it" doesn't seem like a strong argument, emergent behavior and whatnot.
Hell, the fact we don't know what consciousness is means that some LLM systems (or trees! or planets!) might already have it, and we'd be none the wiser. Not that I believe this, but as these systems get larger and more sophisticated it's an increasingly concerning proposition that we have no way to validate either way.
IMO consciousness is just a human-centric ego concept, nothing to do with reality. It's just too much to consider the possibility that "the brain just take the decision." Most people just want the narrative of "I take the decision; the brain follows", which is just false.
We don't know why do many medicines work, and that doesn't stop us from using them to good results.
AI will give us good results, in fact it already has. But like medicine, there will be side effects that will be felt by a lot of people, short-term and long-term. But AGI is to software what a miracle drug is to medicine. Maybe it exists, but we haven't the faintest clue on how to go about it.
[flagged]
[flagged]
[flagged]
[dead]
[dead]
This is copiummaxxing.