Nitpick

What it is

The convention of prefixing a comment with “nit:” exists because reviewers know the difference between a preference and a problem. The prefix is a promise: this is not worth blocking on.

Why AI reviewers produce so many

Nitpicks are the easiest findings to generate. They need no understanding of intent, no knowledge of the wider codebase, and they are almost never wrong in a way anyone can point at. A tool trying to look useful in its first week will produce them by the dozen, and a vendor demo full of them looks thorough.

The cost lands later. Nitpicks are indistinguishable from real findings at a glance, so they teach the team to skim — which is alert fatigue, arriving by the most avoidable route.

Where they belong

Almost every nitpick is a job for a formatter or a linter, running before the pull request exists. If a rule can be checked deterministically, it should be enforced deterministically and silently, not raised as an opinion by a model in the middle of review.

That division of labour is what the default-quiet standard argues for: the reviewer should be arguing about behaviour, not whitespace.

Common mistakes

• Enabling style categories in an AI reviewer while also running a formatter, so the same issue is raised twice.
• Judging a trial by comment volume.
• Letting nitpicks block, which is how teams learn to ignore the bot entirely.