LGTM fatigue

How to stop rubber-stamping pull requests

"Looks good to me" has become the review. Why rubber-stamping happens when agents write the code, and a practical way to review decisions instead of approving them.

"LGTM" used to mean you looked. Now it often means you trust the tests, the agent, or the person who asked the agent. Everyone knows it, and the approvals keep coming.

Why it happens

Rubber-stamping is not laziness. It is what people do when reviewing properly costs more than they have:

• the change is too big to hold in your head
• the explanation is prose you have to translate back into code
• the author, human or agent, has moved on to the next thing

So the review becomes a check that nothing looks alarming, and the decisions go through unexamined.

Lower the cost of a real review

You will not fix this by asking people to try harder. Make a real review cheaper than a fake one:

1. Smaller units. One concern per review.
2. Decisions first. The author says where the choices are.
3. Evidence beside the claim. No "see line 412". Show line 412.
4. A cheap way to disagree. A comment on the exact line, answered by whoever holds the context.

Where deck fits

deck is that cheaper review, when an agent did the work. The agent shows you its changes as a short deck, one claim at a time with the code lit beside it. Disagreeing is selecting a line and typing. The same agent answers, in the same session, on the same lines.

It is not a gate and it does not replace your PR process. It is the moment before you approve, where you actually decide.

Signs it is working

• Review comments name a decision, not a style choice.
• People can explain a change a week after merging it.
• "Why did we do it this way?" has an answer someone remembers choosing.