Simon Willison: Engineering practices that make coding agents work - The Pragmatic Summit

By Pragmatic Engineer

Categories: Product, Startup

Summary

Simon Willison reveals AI coding agents have crossed a critical trust threshold: Opus 4.5 is now reliable enough that top teams skip code review entirely, relying instead on test-driven development and behavioral verification. The shift from "reading every line" to "trusting the output" happened just 6 months ago—and it's reshaping how productive engineers actually work.

Key Takeaways

  1. Evolution of AI adoption has 5 distinct stages: question-asking → code snippets → agents writing most code → no human code writing → no code reading at all. The transition to skipping code review happened only 3 weeks ago with Strong DM's software factory model.
  2. Opus 4.5/4.6 achieved the trust inflection point ~6 months ago. Simon now confidently delegates complex tasks (JSON API with database pagination) without review, getting reliable outputs consistently on repeat problems.
  3. Red-green test-driven development is the core engineering practice enabling code-free workflows. Tests replace code review as the verification mechanism—agents prove correctness through passing tests rather than human inspection.
  4. Real-time AI coding shows 45-49% performance improvements in production code. Simon shipped a 49% Fibonacci optimization and atom feed features within 30 minutes on his phone using Claude—demonstrating mobile-first AI-assisted development.
  5. Reframe trust model: treat AI agents like external service teams. You don't read their source code—you test their API contracts and behavior. This psychological shift from code reviewer to product consumer makes high-trust workflows psychologically sustainable.

Topics

Transcript Excerpt

Um, thank you for joining us today. Uh, as Sammy said, my name is Eric. Uh, I lead infrastructure and security at Statig. Uh, today I get the pleasure of chatting with Simon here, uh, about coding agents. Um, so for those who do not know Simon, uh, Simon is an active contributor to the open source community, maintains hundreds, thousands, >> it's hundreds, there's a thousand repos, but only hundreds of them are maintained. >> Okay. Okay. There we go. Hundreds of repos maintained. Um is the creat...