There's a framework making the rounds in engineering circles — Dan Shapiro's “Five Levels of AI-Assisted Programming,” which maps the progression from basic code completion to the fully autonomous “dark factory” where AI ships software without human intervention on the code.
We built an entire brand around it. Not because it's trendy — because it's the clearest articulation of where software engineering is heading, and we want to be transparent about where we are on that path.
Where We Are: Level 3
Today, Dark Agent Factory operates at Level 3 — what Shapiro calls “Agent-Assisted Development.” In practice, this means Claude Code is our primary engineering tool. We write terminal briefs (detailed specs), and AI agents implement the features, write the tests, and open pull requests.
The evidence: in one 7-day sprint (February 5-12, 2026), we shipped 18 PRs on the AdaptiveTest platform. That included security hardening, audit logging, billing integration, error monitoring, and rate limiting. One engineer. 532 tests. Zero production incidents.
But here's the honest part: we still review every PR. We still make the architecture decisions. We still direct priorities. The agent builds; the human engineers the process.
What Level 4 Looks Like For Us
Level 4 is spec-driven production — write a product requirement, and the AI implements it completely. We're not there yet, but we can see the path:
First, auto-triage. When Sentry catches an error, an agent should analyze the stack trace, write a fix, run the tests, and open a PR — all before we even check our notifications.
Second, spec-to-deploy. Hand the agent a full product spec (not just a terminal brief), and it handles everything: code, tests, migration, deployment, and verification.
Third, AI-generated test suites. The agent doesn't just write code — it writes comprehensive tests for that code, achieving 90%+ coverage autonomously.
Why Level 5 Matters
Level 5 — the dark factory — is the end state. Business requirements go in, deployed software comes out. The human focuses on strategy, client relationships, and quality oversight. The agents handle everything else.
Nobody is there yet. We believe the path exists, and we're walking it publicly. Every month, we'll publish a “Level Up Report” with real metrics: what we automated, what we still do manually, and what's next.
Follow along. The lights are going off.