The dark factory is real, most developers are getting slower, and your org chart is the bottleneck (plus 5 prompts to get from level 2 to 5)

Three engineers at StrongDM run a software factory where no human writes code and no human reviews code. The system takes specifications written in markdown, builds the software, tests it against behavioral scenarios, and produces shippable artifacts. The humans approve outcomes.

Meanwhile, 90% of Claude Code’s codebase was written by Claude Code itself. Internal teams at Anthropic describe operating at 70–90% AI-written code, with some workflows approaching nearly all implementation generated by Claude Code. Boris Cherny, who leads the project, hasn’t personally written code in over two months.

And at the same time, in one rigorous randomized controlled trial, experienced open-source developers working in codebases they already knew completed tasks 19% slower when using AI tools. Not faster. Slower. And they don’t know it. They predicted AI would make them 24% faster, and after the study they still believed it had made them 20% faster. They were wrong about the direction, let alone the magnitude.

The frontier teams aren’t just using better AI tools. They’ve rebuilt everything around a fundamentally different workflow — one where the bottleneck has moved from how fast you can write code to how precisely you can describe what should exist, and where the organizational structures designed for humans writing code have become friction rather than support. Most developers who call themselves “AI-native” are stuck reading every diff the AI produces. The teams running dark factories have moved past reading code entirely. The gap between those two realities is enormous, and closing it requires changes that go far beyond installing a better tool.

Here’s what’s inside:

• How the dark factory actually works. StrongDM’s three-person team, the spec-driven workflow, and the scenario architecture that prevents agents from gaming their own tests.

• The machines building themselves. 90% of Claude Code written by Claude Code, Codex 5.3 instrumental in creating itself, and what a closed feedback loop means.

• The honest distance. Why most developers are measurably slower with AI tools — and why they don’t know it.

• The org chart problem. What happens to sprint planning, code review, and engineering management when coordination becomes friction.

• The legacy problem. Why you can’t dark-factory your way through existing systems, and what the migration path actually looks like.

• The talent reckoning. Junior developer employment collapsing, the career ladder hollowing out, and the rising bar for what “good engineer” means.

• The shape of the new org. Tiny teams generating $100M+ in revenue and what the org chart looks like when agents do the implementation.

• The tension that won’t resolve. The frontier is further ahead and the middle is further behind than anyone wants to admit.

The dark factory doesn’t run on more engineers. It runs on engineers who can think clearly about what should exist, describe it precisely enough that machines can build it, and evaluate whether what got built actually serves the humans it was built for. That’s a fundamentally different job than the one most engineers were hired to do.