The key skill for coding agents: instructing and verifying changes

Bulletin. Simon Willison, writing on his personal site simonwillison.net on August 22, published a short piece titled “More than just code review,” originally headlined “The key skill for coding agents: instructing and verifying changes.”

Willison’s argument, as summarized in the piece: the central skill for productive use of coding agents is not the ability to write code oneself but the ability to confidently instruct an agent on the changes needed, and then confidently verify that those changes were applied correctly.

He notes that verification does not require reviewing every line of generated code. Line-by-line review, he states, has never been the most effective way to validate a change to a piece of software. Other verification methods exist and can serve the same purpose.

The piece is brief, framed as a single observation about where the bottleneck in agent-assisted coding now sits. Willison does not present it as new to his broader body of writing on coding agents, but as a distillation of a point about workflow rather than model capability.

No further detail on specific verification methods, tooling, or examples is given in the material provided. The post carries Willison’s standard tag set linking to his ongoing commentary on AI-assisted software development, according to the source.