Code review after code gets cheap: evidence and blast radius matter more than diff size

O'Reilly Radar argues that reviewing agent-generated code should rely on agreed behavior, evidence, and blast-radius checks rather than reading every changed line.

Using a fictional refund-service example, the article says a small diff can add payment retries that risk duplicate refunds.

It recommends tracing changed behavior in both directions between requirements and implementation, and checking that tests observe the agreed boundaries.