> The factory of the future isn’t a coding agent, an IDE plugin, or a model API. It’s a full-stack service that accepts a spec from a brand and delivers running software continuously. That means model orchestration, code creation, hosting, implementation, testing, deployment, monitoring, maintenance, evolution. End to end.
> Nobody has built this yet. But the pieces are falling into place fast. A factory isn’t a demo you vibe code in a weekend. It’s harnesses and production ownership — deep, compounding infrastructure that becomes exponentially harder to compete with once it’s running.
Definitely a direction that I'm sure people are working towards, but I think this underestimates how crucial human expertise is in building software.
Basically, I don't think software will ever become as commoditized as sneakers are.
At a complex level of requirements (which will increasingly be the case with future software), software engineering becomes about architecture and constraint satisfaction where finding a really-out-of-the-box idea which you build into the software at the ground up that makes everything else possible becomes disproportionately important.
I think people with a combination of technical & market expertise directly working on the software with full context of the project will continue to beat "brands" outsourcing software entirely to "factories".
(That said, 100% automated "factories" will probably be relevant in automatically fast-following competitors or rapidly catching up to existing moats. This could make swathes of software un-monetizable, sort of like drug development without patents. You could spend a lot of effort finding a software design only for people to copy your "formula" and produce "generics" without any of your R&D effort.)
I wonder if this will be true because we're interested in software for humans and the AIs are not as good at understanding humans. The recent news that AIs select for their own resumes over the resumes of humans and other AIs makes me wonder if the AIs will be good at producing software for their own "needs", but not for humans whom they don't "understand" as well.
Applying a production paradigm rooted in the patterns of the past (factories) is foolish. Factories automate repeatable process, where each repetition produces a unit of value. There is nothing to repeat in software.
The configuration capability of factories is only there such that you can then go on to run the repeatable process. No one would build configurable factories if you were only going to run one unit.
With that out the window, the factory analogue breaks down entirely. Think of something else.
So the argument is that the brand will understand the problem domain well enough to define it so an outsourced software factory can build it. That factory will also run it and maintain it.
I've seen this in the consulting world with long term relationships between a brand and a consulting company, where the consulting company is the technology partner.
I don't think there's anything to stop that from happening with agents; it's just a different means of producing software.
So the central claim is this:
> The factory of the future isn’t a coding agent, an IDE plugin, or a model API. It’s a full-stack service that accepts a spec from a brand and delivers running software continuously. That means model orchestration, code creation, hosting, implementation, testing, deployment, monitoring, maintenance, evolution. End to end.
> Nobody has built this yet. But the pieces are falling into place fast. A factory isn’t a demo you vibe code in a weekend. It’s harnesses and production ownership — deep, compounding infrastructure that becomes exponentially harder to compete with once it’s running.
Definitely a direction that I'm sure people are working towards, but I think this underestimates how crucial human expertise is in building software.
Basically, I don't think software will ever become as commoditized as sneakers are.
At a complex level of requirements (which will increasingly be the case with future software), software engineering becomes about architecture and constraint satisfaction where finding a really-out-of-the-box idea which you build into the software at the ground up that makes everything else possible becomes disproportionately important.
I think people with a combination of technical & market expertise directly working on the software with full context of the project will continue to beat "brands" outsourcing software entirely to "factories".
(That said, 100% automated "factories" will probably be relevant in automatically fast-following competitors or rapidly catching up to existing moats. This could make swathes of software un-monetizable, sort of like drug development without patents. You could spend a lot of effort finding a software design only for people to copy your "formula" and produce "generics" without any of your R&D effort.)
I wonder if this will be true because we're interested in software for humans and the AIs are not as good at understanding humans. The recent news that AIs select for their own resumes over the resumes of humans and other AIs makes me wonder if the AIs will be good at producing software for their own "needs", but not for humans whom they don't "understand" as well.
Applying a production paradigm rooted in the patterns of the past (factories) is foolish. Factories automate repeatable process, where each repetition produces a unit of value. There is nothing to repeat in software.
The configuration capability of factories is only there such that you can then go on to run the repeatable process. No one would build configurable factories if you were only going to run one unit.
With that out the window, the factory analogue breaks down entirely. Think of something else.
This appears to be written by somebody who makes shoes, not software.
Haven't you heard? Shoe companies are pivoting to AI
Gentlemen, this is Democracy manifest
So the argument is that the brand will understand the problem domain well enough to define it so an outsourced software factory can build it. That factory will also run it and maintain it.
I've seen this in the consulting world with long term relationships between a brand and a consulting company, where the consulting company is the technology partner.
I don't think there's anything to stop that from happening with agents; it's just a different means of producing software.