how much of the agent's code are you actually reading vs just approving

A developer using coding agents finds that the bottleneck has shifted from code writing to code review, specifically trusting the agent's output. A helpful strategy involves having the agent write a plan first, which is easier to review than 200 lines of code. Routing data fetches through "context dev" to ensure current documentation is used also improved plan quality. The developer asks how others manage this review process and if they ever reach a point of trusting the agent enough to stop reviewing every line.

This report uniquely highlights a shift in the developer bottleneck from writing code to reviewing agent-generated code, unlike earlier discussions focused on code generation itself.