A short piece — posted as a note rather than a full article on Simon Willison’s weblog on 22 August 2026 — making a single argument about what productive work with an AI Coding Agent demands of the developer directing it. Willison puts the key skill in two halves that have to hold together: being able to confidently instruct the agent on how to make changes, and then being able to confidently verify that those changes have been applied in the correct way.
The second half is where the note pushes back on a common assumption. Reviewing every line of code an agent has written is one way of reaching that confidence, but Willison holds that it is not the only one — and that eyeballing every line has never been the most effective way to validate a change to a piece of software in the first place.
Key Points
- Willison asserts that the key skill required to make productive use of coding agents is being able to confidently instruct them on how to make changes and then confidently verify that those changes have been applied in the correct way. The note offers this as his own position and presents no supporting evidence for it.
- Verifying an agent’s work sometimes involves reviewing every line of code it has written, but on his account there are other ways to achieve the same goal.
- He argues that eyeballing every line of code has never been the most effective way to validate a change to a piece of software — stated as a general claim about software changes, not one confined to agent-written code.