Summary
Cole Medin breaks down Dan Shapiro’s “five levels of AI coding” framework — from “spicy autocomplete” to the “dark factory” — to help developers locate themselves on the autonomy spectrum and decide where to aim next. Borrowing the structure from the Society of Automotive Engineers’ self-driving car taxonomy, the framework maps AI coding from level zero (AI as a smarter Stack Overflow that never writes a line) through level five (a fully autonomous software factory with no human involvement). Medin argues most developers are stuck at level two — the “junior developer” stage — and should be pushing toward level three, where the AI writes the majority of code while a human directs, reviews, and validates.
A central theme is healthy skepticism about the “dark factory” model despite its growing popularity. While companies like StrongDM are reportedly running fully autonomous AI coding pipelines even for production banking code, Medin — who hasn’t written a single line of code himself in over a year — emphasizes that reaching level five requires a robust system for planning, implementation, and validation that most teams haven’t built yet. The real risk is organizations jumping to level five tooling without the scaffolding to catch AI errors before they compound.
The video also spotlights Sonar’s acquisition of Guitar, an AI code review tool that automatically identifies and fixes issues — including SQL injection vulnerabilities — and validates every fix against CI pipelines. Sonar cites a 44% reduction in production outages from AI-generated code among its users, offering a concrete benchmark for teams evaluating their AI coding safety nets.
📺 Source: Cole Medin · Published July 03, 2026
🏷️ Format: Opinion Editorial
