Willison publishes agentic engineering patterns guide

Willison's guide arrives at the exact moment the industry is debating what "software engineering with AI" actually means — and he's giving it a name, a definition, and a methodology.

• –Defines agentic engineering around one core principle: agents run tools in a loop, and code execution is what separates them from chatbots
• –The human role shifts from writing code to specifying problems precisely, verifying outputs rigorously, and updating instructions as agents iterate
• –Code is now "inexpensive" — the new scarcity is domain expertise and the judgment to navigate architectural tradeoffs
• –Recommends test-first development with agents so they can verify their own output without constant human supervision
• –The guide is designed to evolve over time — living documentation, not a frozen snapshot — covering testing, code understanding, and practical techniques