Why it matters for agencies
A second pair of eyes catches mistakes that the author has stopped being able to see. Reviews also spread knowledge of the codebase and keep standards consistent, but they stall when pull requests are large or reviewers are busy.
How to do it or work it out
Keep changes small and described clearly. Reviewers should check that the change does what it says, handles errors, does not weaken security, and is readable enough that the next person can change it.
Agree what a machine can check so people do not spend attention on it: formatting, obvious bugs and missing tests. Save human review for design and intent.
Review in a sensible order: does the change do what the task asked, is it safe, is it tested, and only then is it readable. Style arguments come last and are best settled by a formatter.
A worked example
A pull request changes 300 lines at the end of the day. An automatic first pass flags a missing check on user input and a duplicated function. The author fixes both before a teammate looks, and the human review can focus on whether the approach is right. The human reviewer then asks why the duplicated function was not reused, the author consolidates it, and the pull request merges after one round of review instead of three.
Common mistakes
- Reviewing huge changes. Past a few hundred lines reviewers skim, so ask for smaller pull requests.
- Treating review as a verdict on the author. Comments are about the change, and good teams word them as questions or suggestions.
- Treating an AI first pass as the review. It can miss design problems and business rules, so a person still decides whether to merge.