After six months of slow deploys and bottlenecked PRs, we stripped everything down. No more synchronous "let's pair review this". No more waiting for the one person who knows the subsystem.

What changed: reviewers write detailed, actionable comments with context. if you say "this is inefficient", you explain why and link to the relevant perf test. if you're unsure, you say so instead of rubber-stamping. authors respond in writing, not in a meeting where half the context gets lost.

The key part was killing the "you need approval from 2 people" rule and replacing it with "you need responses from the relevant experts". a logging service only needs the observability person, not the entire team.

we also started requiring PRs to have a one-sentence summary of what changed and why. sounds dumb, but it's the difference between "refactor database layer" and "batch insert 1000 events instead of individual writes. benchmarks show 60% reduction in lock contention under 5k/sec load". reviewers immediately know what to care about.

downside: it takes discipline. people still want to skip writing explanations. and now you can't hide behind "the reviewer should have caught it."

but our deploy frequency tripled. that seems worth it.