Karat proposes agent PR interview with three planted bugs
The author argues that as AI coding agents make generating code cheap, technical interviews should shift from build rounds to review rounds centered on evaluating an agent's pull request containing planted bugs. This approach tests a senior engineer's scarce skills of spotting wrong assumptions, verification gaps, and removed safeguards rather than merely measuring output volume. The proposed 60-minute interview format includes reviewing the PR, running tests, and discussing whether to merge it and how the agent should have been briefed. Open article →