The “Senior Who Turns 50 Lines Into One” Is Now an Agent
Encode review taste as rules and examples, then let an agent apply the same compressions a strong reviewer would.
Every team has a senior who collapses a messy function into something obvious. That taste used to live only in their head and their PR comments.
You can encode a large share of it: naming rules, “prefer stdlib,” “no drive-by refactors,” “fail loud,” example before/after pairs. An agent with those constraints starts making the same class of compressions — not perfectly, but usefully on the boring 70%.
The point is not replacing seniors. It is multiplying the hours they would have spent on nits so they can spend judgment on design. Put the taste in the repo. Let the agent swing first. Keep humans on the merges that matter.
ニュースレターに登録
最新のAIに関するインサイトをお届けします。