Don't accept AI output without understanding it
It’s easy to let the agent do its thing and move on. You get a working feature, the tests pass, the diff looks reasonable. So you ship it.
But “looks reasonable” is not the same as “I understand what this does.” Especially in areas of the codebase you’re less familiar with, the agent can introduce patterns, dependencies, or logic that you wouldn’t have chosen — and wouldn’t catch without reading carefully.
You get bitten when you fall for it. A subtle bug, a security hole, a dependency you didn’t need. And when it breaks, you’re the one debugging code you never actually read.
Review the code as if you wrote it, because your name is on it. Own your contributions.