The 5 Levels of AI Production
From spicy autocomplete to the dark factory — a framework for understanding where you are and where you're going.
Framework inspired by Dan Shapiro's “Five Levels of AI-Assisted Programming” (January 2026)
Level 0: Manual Coding
Traditional hand-coded software development. No AI assistance.
What it looks like:
- Every line of code written by hand
- Debugging by reading code and adding print statements
- Copy-pasting from Stack Overflow and documentation
Human role:
Everything. The human is the entire production line.
Level 1: Spicy Autocomplete
AI suggests code completions. Developer accepts or rejects inline.
What it looks like:
- Tab-completion on steroids — AI finishes your lines
- Suggestions for function bodies, boilerplate, common patterns
- Still fundamentally human-driven, just faster typing
Human role:
Writer. The human writes code with an AI spell-checker.
Level 2: AI Pair Programmer
AI writes blocks of code in conversation. Developer reviews and integrates.
What it looks like:
- Ask AI to write a function, component, or test
- Chat-based interaction — describe what you want, get code back
- Developer still assembles the pieces and handles architecture
Human role:
Senior developer. The human architects; the AI implements fragments.
Level 3: Agent-Assisted Development
Where We AreAI agents write features end-to-end. Developer reviews PRs and provides direction.
What it looks like:
- Give the agent a terminal brief — it writes the feature, tests, and opens a PR
- Multiple PRs per day from AI agents
- Developer reviews code, catches edge cases, makes architecture calls
- CI/CD handles testing and deployment automatically
Human role:
Code reviewer and architect. The human directs; agents build.
Our evidence:
18 PRs in 7 days. 660+ automated tests. Two AI services deployed. Claude Code as primary engineering tool.
Level 4: Spec-Driven Production
Developer writes specs. AI handles all implementation, testing, and deployment.
What it looks like:
- Write a product requirement document → AI implements it completely
- AI agents auto-triage bugs, write fixes, deploy patches
- AI generates its own tests, achieving 90%+ coverage autonomously
- Human only intervenes for product decisions and quality audits
Human role:
Product manager. The human defines what to build; agents handle how.
Our roadmap:
Auto-triage Sentry errors → fix PRs. Spec-to-deployment pipeline. AI-generated test suites.
Level 5: The Dark Factory
The VisionFully autonomous. AI receives business goals and ships software end-to-end.
What it looks like:
- Business requirement in → deployed, monitored software out
- Self-healing systems that detect, diagnose, and fix production issues
- Multi-agent orchestration — specialized agents for frontend, backend, testing, DevOps, security
- Continuous improvement without human intervention on code
Human role:
Business strategist. The human sets direction and manages client relationships. That's it.
Reality check:
Nobody is here yet. We believe the path exists, and we're walking it publicly.
Dark Agent Factory: Level 3, Moving to 4
Here's the evidence — real numbers from a real production sprint.
What Level 4 Looks Like for Us
- Auto-triage Sentry errors → fix PRs
- Spec-to-deployment pipeline
- AI-generated test suites
The Dark Factory
Business requirement in → deployed software out. Self-healing systems. Multi-agent orchestration. Continuous improvement without human intervention on code.
We're not there yet. Nobody is. But we're building toward it publicly. Every month, we share real metrics, real lessons, and real progress.