StrongDM introduces the idea of a Digital Twin Universe to validate end-to-end “scenarios” for software built (they prefer the term “grown”) in “dark factories”1 by agents.
So if you want to run a software Dark Factory, you need specs as an input—StrongDM calls this the “Seed”—and those can be generated by agents (again I’m using the /prd and /ralph Skills); the other thing you need is an ability to define and run Scenarios—these start as user stories, but the idea is you try to cover anything a user could conceivably do with your software to surface bugs.
I have a new repo where I’m using the Ralph Wiggum /prd and /ralph Skills in Claude Code to generate detailed PRD and task files that can be fed to Ralph or Gastown or some other agent work orchestrator to replicate various well-documented APIs. Basically you point Claude at public API documentation and say “generate a PRD for a shallow clone.” By “shallow” I mean that we won’t replicate the data infrastructure; we’ll stub things out or return dummy data from endpoints.
The whole StrongDM website is full of fascinating ideas and techniques, it’s worth looking through. Simon Willison also has a nice writeup: https://simonwillison.net/2026/Feb/7/software-factory/
Footnotes
The factories are dark because they’re manned by machines; the machines don’t need light to work. Eerie but maybe prescient?↩︎