Used to play quite a bit with Mastra (not Factory) and it was a very pleasant experience. Cool launch, congrats!
Couple of questions:
1. Did you experiment with not having a separate design/plan phase? From my experience agents usually don’t really know what the plan is until they start editing files and executing tests (in other words, implementing), so the initial plan is usually not great and changing it mid-flight is overhead. I also find reviewing tested code to be much easier than reviewing plans/designs, which are often full of ungrounded assumptions.
2. Is this using your own custom coding agent harness over the raw LLM endpoints users provide? If so, did you consider having it integrate with existing harnesses (e.g. Claude Code, Antigravity, etc.)?
hmm, last year people were mostly building in-app assistants but now it's a much wider variety: ai sres, gtm agents, all kind of internal slack agents, factory type things, healthcare or financial services specific stuff and so on
but it all seems to come down to the same primitives around durable agents and workflows, memory, knowledge, sandboxes, filesystems.
the one exception is actually factory, it's a bit of a departure from what we've done (since it's application on top of mastra primitives vs the primitives themselves).
Unfortunately I think so. Folks like sindresorhus that also closed their repos had very similar reasoning to ours. The number of PRs of marginal value has skyrocketed at the same time as it's become much easier to do those quickly yourself.
we've been doing open source for ~10yrs so maybe we have a bit of nostalgia for the way things have been?
also there are a few folks on our team who we got to know because they showed up in our repo and started contributing, and maybe that sort of serendipity won't happen in the future
Well, maybe. Contributing code is definitely a proxy for caring and knowing your way around. But perspectives on how the software works for people, what use cases it could accommodate and how are ever important. Now we can focus on that and I'd argue it's much more interesting.
Using GitHub issues seems to increase the number of those you get by quite a lot - seems like people have bots that scan public issues and generate PRs for them. If you use something else as your issue tracker and keep GitHub issues empty you stop a decent chunk of low effort automated PRs.
I suspect sticking your public repo somewhere other than GitHub and sharing it in more organic programming communities (discord, forums, your hn bio etc) would further reduce the amount of low effort slop you get.
Yeah i think so! The contribution model of OSS has shifted, as maintainers we want good issues with good reproductions. BUT even this has a new level of slop. Opus really likes making issues that you would never have even prioritized in any sprint ever. Like, it's not wrong but also why should I care about that issue. I think this is the next frontier: we should only fix issues that actually matter.
1. Clone repo A and B + Start OpenCode + Feed a prompt into it
2. Pick up the commit generated by opencode for both repo, and run CI
Another workflow I created is: every day at 3AM, check that the documentation of the website is up to date, if yes exit, if not update the documentation. Pass the updated documentation to an adversarial LLM and if it pass, merge it - otherwise loop with the feedback from the adversarial one.
You have a concept of trigger, which can be either a webhook, or a github PR or a github issue, or just a cron expression. Or just someone manually clicking.
There is a concept of secrets and config, so the user can create their own secrets to start codex-cloud, or opencode, or making custom API call.
There is also the concept of manual gate, so you can wait for an human to actually click a button.
Finally, it supports running containers. Useful if you want to call an obscure API from your workflow, you don't have to wait for the platform owner to create a step for your API, but you can just run a container into it.
This needs more visuals - the kinds of people using this may not be software developers, and it seems just like a command line too at first glance unless you watch the video.
We enable choosing / customizing this based on org/user preferences. By default, the identity is the GitHub App "Mastra Platform" but configuring with a personal access token enables the agent loop to open PRs on the behalf of users.
Open to input on what you have found is best, but in general a belief we have is to not be too dogmatic one way or the other and enable users to choose.
> For us Factory is a configurable agentic kanban board that's @-taggable in Slack.
Software factories are not a real architecture. They are just AI agents working on the same project. GitHub and similar platforms are already ideal for this. Requirements go in as issues, agents open PRs, someone reviews and merges them.
We used to call this software development.
The only "new" part is that now nobody complains at standups!
The reality is that making a quality product takes a long time, even when relying heavily on generative AI. Most people don't have the time or desire to remake something they don't pay a lot (or anything) for that works well.
i'm always curious how the workflow with these factories work. like are you basically just giving it tickets and letting the ai use it's "judgement" on how to implement them? do you essentially say something like "we need X" and the agents do the rest? or is it somehow more interactive? and what about verification, particularly on the UX side of the end result?
you can customize this a lot based on your needs. You can create skills for each column in a factory board, select different models, and steer the agents as they are running.
Depending on how you have it set up, your agent will stop for you to approve plans, answer questions, etc.
In our factory, low risk/low complexity items are more automated while higher risk/higher comlpexity items need more supervision.
Mastra Factory provides fully working preview environments in a sandbox so you can quickly test UX/UI changes in your app.
thanks. my fear with the “dark factory” so to speak is that you lose the aspect of interacting with the software during the development process which can both surface ux issues immediately and also discover “pivots” or other inspirations as you iterate. preview environments probably help, i’m curious if you’ve encountered these issues and if so how you solve for them.
Used to play quite a bit with Mastra (not Factory) and it was a very pleasant experience. Cool launch, congrats!
Couple of questions:
1. Did you experiment with not having a separate design/plan phase? From my experience agents usually don’t really know what the plan is until they start editing files and executing tests (in other words, implementing), so the initial plan is usually not great and changing it mid-flight is overhead. I also find reviewing tested code to be much easier than reviewing plans/designs, which are often full of ungrounded assumptions.
2. Is this using your own custom coding agent harness over the raw LLM endpoints users provide? If so, did you consider having it integrate with existing harnesses (e.g. Claude Code, Antigravity, etc.)?
You know you reach human level intelligence when you hit human-like constraints.
"I wont know until I look, and it'll take a few iterations to get it right"
Links:
https://factory.mastra.ai
https://github.com/mastra-ai/template-factory
thanks for dropping these in
What have you learned from people who use Mastra today versus what you/they were doing in the beginning?
hmm, last year people were mostly building in-app assistants but now it's a much wider variety: ai sres, gtm agents, all kind of internal slack agents, factory type things, healthcare or financial services specific stuff and so on
but it all seems to come down to the same primitives around durable agents and workflows, memory, knowledge, sandboxes, filesystems.
the one exception is actually factory, it's a bit of a departure from what we've done (since it's application on top of mastra primitives vs the primitives themselves).
> closed our repo to outside contributors
I've seen this pattern happening in other open-source repos. Do you think it's going to be the norm now?
Unfortunately I think so. Folks like sindresorhus that also closed their repos had very similar reasoning to ours. The number of PRs of marginal value has skyrocketed at the same time as it's become much easier to do those quickly yourself.
I don’t find it unfortunate at all. This makes it much easier to accommodate requests the right way.
we've been doing open source for ~10yrs so maybe we have a bit of nostalgia for the way things have been?
also there are a few folks on our team who we got to know because they showed up in our repo and started contributing, and maybe that sort of serendipity won't happen in the future
Well, maybe. Contributing code is definitely a proxy for caring and knowing your way around. But perspectives on how the software works for people, what use cases it could accommodate and how are ever important. Now we can focus on that and I'd argue it's much more interesting.
i think the big challenge is it changes the open source contribution model, but this is why we typically give contribution credit for good issues now.
Using GitHub issues seems to increase the number of those you get by quite a lot - seems like people have bots that scan public issues and generate PRs for them. If you use something else as your issue tracker and keep GitHub issues empty you stop a decent chunk of low effort automated PRs.
I suspect sticking your public repo somewhere other than GitHub and sharing it in more organic programming communities (discord, forums, your hn bio etc) would further reduce the amount of low effort slop you get.
Yeah i think so! The contribution model of OSS has shifted, as maintainers we want good issues with good reproductions. BUT even this has a new level of slop. Opus really likes making issues that you would never have even prioritized in any sprint ever. Like, it's not wrong but also why should I care about that issue. I think this is the next frontier: we should only fix issues that actually matter.
So... In some ways getting the telemetry of the real use (and issue) is more valuable than the fix.
Exactly!
I build something similar, based on similar ideas: https://factory.redbeardlab.com
There is a [language](https://redbeardlab.github.io/dotFactory/) that let express complex workflows.
The workflow can be pretty much anything, eg:
1. Clone repo A and B + Start OpenCode + Feed a prompt into it
2. Pick up the commit generated by opencode for both repo, and run CI
Another workflow I created is: every day at 3AM, check that the documentation of the website is up to date, if yes exit, if not update the documentation. Pass the updated documentation to an adversarial LLM and if it pass, merge it - otherwise loop with the feedback from the adversarial one.
You have a concept of trigger, which can be either a webhook, or a github PR or a github issue, or just a cron expression. Or just someone manually clicking.
There is a concept of secrets and config, so the user can create their own secrets to start codex-cloud, or opencode, or making custom API call.
There is also the concept of manual gate, so you can wait for an human to actually click a button.
Finally, it supports running containers. Useful if you want to call an obscure API from your workflow, you don't have to wait for the platform owner to create a step for your API, but you can just run a container into it.
Thanks for sharing!
Legendary and congratulations on the launch!
thank you
This needs more visuals - the kinds of people using this may not be software developers, and it seems just like a command line too at first glance unless you watch the video.
Thanks for watching the video -- assume you're referring to the one on the blog post? https://mastra.ai/blog/announcing-new-capabilities-in-mastra...
Actually it was this one... https://m.youtube.com/watch?v=iMA-Xkhj7fU
Just to clarify, as I couldn’t confirm this from the docs. Do agents run with their own identity or is it assuming engineer github identity?
We enable choosing / customizing this based on org/user preferences. By default, the identity is the GitHub App "Mastra Platform" but configuring with a personal access token enables the agent loop to open PRs on the behalf of users.
Open to input on what you have found is best, but in general a belief we have is to not be too dogmatic one way or the other and enable users to choose.
Great. How’d I do that? I read the docs and also asked the agent the same question but it said that it was not possible. Thank you.
> For us Factory is a configurable agentic kanban board that's @-taggable in Slack.
Software factories are not a real architecture. They are just AI agents working on the same project. GitHub and similar platforms are already ideal for this. Requirements go in as issues, agents open PRs, someone reviews and merges them.
We used to call this software development.
The only "new" part is that now nobody complains at standups!
Excited to try. Do you have a video demoing how you use it (or just best practices for getting up and running well on my team)?
we did a workshop on how we use it about a month ago: https://www.youtube.com/live/LzZ_FptHWqs?si=o8Dj3828636_wMG5
we will be posting a shorter demo/best practices video to our youtube soon as well!
but you are going to disappear soon swallowed buy some guyz working for a guy called CLAUDE
The reality is that making a quality product takes a long time, even when relying heavily on generative AI. Most people don't have the time or desire to remake something they don't pay a lot (or anything) for that works well.
we have been hearing this for the last 15 months
rumors of our demise have been greatly exaggerated...
at the end of the day it's open source and it's working well for us!
this is cool!
Thanks!
thank you
nice project
Thanks!
i'm always curious how the workflow with these factories work. like are you basically just giving it tickets and letting the ai use it's "judgement" on how to implement them? do you essentially say something like "we need X" and the agents do the rest? or is it somehow more interactive? and what about verification, particularly on the UX side of the end result?
this is a good question
you can customize this a lot based on your needs. You can create skills for each column in a factory board, select different models, and steer the agents as they are running.
Depending on how you have it set up, your agent will stop for you to approve plans, answer questions, etc.
In our factory, low risk/low complexity items are more automated while higher risk/higher comlpexity items need more supervision.
Mastra Factory provides fully working preview environments in a sandbox so you can quickly test UX/UI changes in your app.
thanks. my fear with the “dark factory” so to speak is that you lose the aspect of interacting with the software during the development process which can both surface ux issues immediately and also discover “pivots” or other inspirations as you iterate. preview environments probably help, i’m curious if you’ve encountered these issues and if so how you solve for them.
see https://yegge.ai/wyvern as an (imo) pathological case of this.
(not a criticism, just curious :))