@ahmetcantryk
7 min read
How to Run Effective Code Reviews
Effective code review means hunting for risk, not nits. Here's how small PRs, fast turnaround, tone, and skeptical review of AI-generated PRs work in 2026.
Tag
Effective code review means hunting for risk, not nits. Here's how small PRs, fast turnaround, tone, and skeptical review of AI-generated PRs work in 2026.
METR's RCT found developers were 19% slower with AI while feeling 20% faster. July 2026: the evidence, the conflicting data, and how to measure your own speed.
curl's genuine-report rate cratered from over 15% to under 5% in 2025 as AI slop flooded HackerOne. Here's the signal-vs-noise fight facing maintainers.
Write unit tests that catch real bugs: name for behavior, use Arrange-Act-Assert, mock only boundaries, and skip brittle implementation-detail tests.
With AI writing half the code, clean code principles matter more, not less. A checklist to audit naming, function size, comments, and tests in 30 seconds per PR.
How to refactor legacy code safely in the AI era: pin behavior with characterization tests, ship tiny reversible steps, and use the strangler fig pattern.