Getting to the Dark Software Factory
How to build a dark software factory in practice — machine-derivable intent, lanes and qualifiers, the forager that cherry-picks work, and why attention becomes engineering's real constraint.
Considered writing on AI-assisted development, the dark software factory, Claude Code, and building human. By Stephan Schmidt.
How to build a dark software factory in practice — machine-derivable intent, lanes and qualifiers, the forager that cherry-picks work, and why attention becomes engineering's real constraint.
You can run a dark software factory on your codebase today. Six categories of work already ship autonomously — what works now, what's next, and how to start this week.
Level 6 is a self-optimizing factory that tunes its own harness. Level 7 is a factory that sources its own intent by talking to customers and reading telemetry. Each level pulls more meta-work inside the machine. Here is what each one unlocks, what it still cannot do, and where the product human goes from here.
The dark software factory is the category for fully autonomous AI software delivery: agents write, test, and ship code while humans stay on intent, policy, and outcome review. Origin, five levels, StrongDM example, critiques, and where humans belong in the stack.