Software factories are pretty controversial here lately...
Some pretty big names think they are just a fad or that they’re not even a real thing.
They say you should focus on good engineering and deterministic outcomes like tests, evals, and CI, and the rest will take care of itself. (The first part is of course true)
As long as developers can ship fast with agents and work can be verified, that’s enough to take advantage of everything AI can do for engineering.
As long as these are true, they say, you don’t need a software factory.
Then there’s the other side of the argument that says that software factories are very much a real thing. And there’s lots of proof too.
We have plenty of examples of companies that built them that are now shipping faster than they ever have. See Shopify for a perfect example. They have thousands of engineers. But thousands of PRs a week are opened by their agents. That’s all possible because of their elite software factory.
So who is right?
Well, both, in their own ways.
A lot of this debate comes down to who your “customer” is. For the “software factory isn’t real” crowd, I would bet their customer is either just themselves, a very small team, or an extremely AI-native team. Or they don’t realize what they have is actually a software factory. Or we’re just disagreeing on the term.
Let’s imagine for a minute that what this camp has created are in fact software factories. They’re the thin, easier kind. They don’t have to deal with the scale issues that we’ll get to like context management, governance, and orchestration.
This isn’t a dig at them, by the way. I’m sure they’re shipping super fast and we can definitely all learn from them.
But once you come down from the very best engineering teams, you meet the other kind of customer.
For the “software factories are real” side, the customer in question is a monolithic enterprise with 3000 developers. They’re trying to significantly raise the floor of their engineering team, not the ceiling. There’s a massive gap in AI adoption internally. Some of their developers are resisting AI altogether while others have automated 95% of their work. The job there is just to get everyone working together.
This camp has completely different things to worry about when building their software factory. They’re making a BIG change.
When you make a big change inside an organization at that size, you actually have to make changes in three different things: technology, people, and process.
(I would argue that in smaller or younger teams, people and process are much easier to align once you have the technology in place.)
But in larger teams, technology is now the easy part. People and process, though, good luck!
And that’s exactly where software factories come in. Enterprises aren’t just building software factories. They’re building a brand new way for their company to work together with agents (people and process). For them, a software factory means everything they need to put in place around agents like governance and human-in-the-loop. AI infrastructure with a harness which might be enough for a small team isn’t enough for them.
They still have stand-ups, waterfall-style development, and code written in COBOL.
Giving every developer Claude Code and lots of tokens will cause more chaos than its worth.
So their software factories are real in the sense that they need to cover more people and process.
Let’s say you have a one-line config change that needs to happen. In a large team, that change has to go through multiple levels of approval. It might take months. That’s a process problem and it can be solved with a software factory, if the team fully commits to the new model.
The list of issues you’d reach at scale is long, but here’s just a few.
For 3000 developers, how do you solve:
Context: What context do you have? How do you deliver it effectively to agents?
Governance: What can agents touch/not touch? How do you govern agents with those rules?
Orchestration: Which parts remain deterministic vs agentic? How do hand-offs, retries work?
Human-in-the-loop: When and how do humans get involved to keep work moving?
AI Registry: What agents, skills, plugins and MCP servers exist? Who owns them?
Metrics: Is the factory working?
Integrations: Can the agents interact safely with our platform tools?
Working through all these problems of scale through a software factory gives large teams a chance to turn over a new page and start fresh. They can throw away almost everything about the way they work and say “We’re starting over”.
Gathering everyone around a platform that facilitates that, that is both flexible enough to grow as the initiative grows but also strict enough that it calms the chaos, will let these huge companies benefit fully from agentic SDLC. But it has to be a full commitment with an agreed-upon place for it to happen: on a software factory platform.
If we put aside the term software factory for a second, we can all agree that we are just automating more things. This time, the entire SDLC.
So do you need a full-blown software factory?
If any of the following are true, you probably do:
You have a large engineering team
You don’t have a team you can dedicate to a custom software factory build
AI adoption in your company is extremely uneven
You need to go through multiple layers of approval to get things shipped
Agents, skills, and workflows aren’t shared across teams
And again, the reason isn’t technology. It’s people and process.
So if you need to change all 3, technology, people, and process, then it IS a software factory you’re looking for.
Thanks for reading!
If you have a comment, please join the discussion about this article on X:






