A conference talk by Simon Willison covering verification patterns for agentic engineering — writing production code with AI coding agents. Key topics include: active refactoring as a code review technique, using agents to prototype and stress-test API designs, maintaining documentation accuracy with diff-based prompts, setting up preview environments via GitHub Actions, reducing blast radius with CSP headers, sandboxed iframes, and WebAssembly, achieving zero tolerance for flaky tests by delegating debugging to agents, and preserving agent transcripts as audit artifacts. The talk draws parallels between large-team software engineering practices (conceptual integrity from The Mythical Man-Month, dark factories) and the challenges individual developers now face when shipping AI-generated code at high velocity.