"This feels off" is a comment that tells the author something is wrong and nothing about what to do next. We banned it, along with a short list of its relatives, and review turned around faster within a fortnight.
The comments we stopped writing
Vague dissatisfaction dressed as feedback: "not sure about this", "could be cleaner", "seems fragile". Every one of these is a real observation the reviewer had, expressed in a form that makes the author guess at the actual concern. Guessing wrong produces a second round of review, which is slower for everyone than writing the concern out properly the first time.
What replaced them
A comment now states the concern and, where possible, the fix or the question that would resolve it:
- "This will throw if the list is empty — handle it or assert it can't be." Not "edge case?"
- "This duplicates the validation in the signup form. Extract it or link to why it needs to differ." Not "duplication".
- "I don't understand why this runs before the auth check — can you add a comment or reorder it?" Not "this seems wrong".
Each version gives the author a next action instead of a feeling to interpret.
This matters more, not less, with agent-written code
A vague comment on a human's code at least reaches someone who can ask a follow-up question in the next standup. A vague comment left for an agent to act on produces a guess, applied automatically, that may address the wrong part of the concern entirely. Specific comments are the only kind that translate reliably regardless of who — or what — wrote the code being reviewed.
Feedback that requires a follow-up conversation to become actionable has not saved anyone any time. It has just moved the cost of clarity from the reviewer to a second round trip.
The harder discipline
Writing a specific comment takes longer than writing a vague one, because it requires the reviewer to finish the thought instead of flagging it. We accept that cost. It is smaller than the cost of a second review cycle, and it is a better use of a senior engineer's attention than restating the same concern twice.
What we watch for
Reviewers who consistently leave vague comments are not being lazy in most cases — they often have not finished diagnosing the problem themselves. Treating a vague comment as a sign the reviewer needs more time, rather than as adequate feedback already given, has changed more review threads than any style guide we have written.