Overview

Dan Shapiro proposes a five-level model for AI-assisted programming evolution, from basic autocomplete to fully autonomous software factories where humans never review AI-generated code. Simon Willison discusses a real team operating at the highest level, building convincing software in months with a tiny team.

View Original

The Breakdown

  • Level 0-2 progression: From spicy autocomplete (GitHub Copilot) to coding intern (boilerplate generation) to junior developer (pair programming with full human review)
  • Level 3-4 shift in human role: Developer level means AI generates most code while humans become full-time reviewers, then engineering team level where humans act as managers collaborating on specs
  • Level 5 “Dark Factory” breakthrough: Complete automation where humans never review AI code, focusing instead on designing systems that prove the software works through extensive testing and tooling
  • Real-world validation approach: Teams operating at Level 5 invest heavily in testing, simulation, and demonstration systems rather than code review to ensure software reliability
  • Dramatic productivity gains: Small teams (under 5 people) with 20+ years experience are building convincing software in just months using this fully autonomous approach