415.tech
AI & tech, from the frontlines of Silicon Valley

Simon Willison: reading every line was never the best way to verify a change

Willison argues that productive use of coding agents rests on two skills — instructing them confidently and confirming the change landed correctly — and that line-by-line review is only one route to the second. That reframes agent code review from a line-count problem into a verification problem, where tests and behavior checks can carry the same confidence. It is a short weblog note rather than a worked argument, and it cites no external sources.

Source: simonwillison.net

Post on XEmail

Eyeballing every line of code has never been the most effective way to validate a chance to a piece of software.

Simon Willison

Why this matters

  • → Tests and behavior checks can verify agent changes as reliably as line review
  • → Shifts focus from reviewing every line to confirming the right outcome occurred
  • → Opens new workflows for faster, confident validation of AI-generated code
Verification, not inspection
Also in this edition