Code review

Code review is the step where a change gets checked before it merges into the main codebase: does it actually work, does it fit the existing patterns, does it introduce bugs, security holes or duplicated logic. Traditionally a second developer reads the diff line by line; increasingly an AI reviewer does a first pass and flags the obvious problems before a human ever looks.

When you are building with AI, review becomes more important, not less. An agent writes code fast and confidently, including the wrong code, so the reviewer is the gate that catches the cut corner, the stub left behind, the test that asserts nothing real. Never trust green as proof: a passing build means the code ran, not that it does what you wanted.

Say you've just had an AI agent ship a new billing flow. Before you merge, you point Claude at the diff with one instruction, find the flaw, don't reassure me, and it surfaces the off-by-one on the proration before a customer ever hits it. To keep the discipline repeatable rather than something you remember on a good day, you turn your review steps into a checklist in Process Street, so every change is walked through the same security, data and dead-code questions. And when the review turns up real problems, you log each one as a follow-up in Linear so nothing gets quietly waved through.

Review the diff with the same scepticism you would apply to a junior's first week. That is how speed stays safe.