Gave Difficult Feedback to a Peer and Nothing Changed? Tell the Second Attempt

A story about giving difficult feedback to a peer is judged on the follow-up. The interviewer asks what changed afterward. A peer doesn’t report to you, so their change is the only result you can claim. Many candidates stop at the conversation. They describe how kind and direct they were. Then the follow-up exposes that nothing measurable happened. The fix is a more honest story. If your first attempt didn’t land, tell it. Then tell the second attempt and what it changed. A strong answer covers five things. Name the problem and what it cost the team. Say what you tried first and why it failed. Describe the second conversation, held in private, with a specific request. Give the result, including the part that didn’t change. Then say what you did about that part. This post follows one Android engineer whose first try changed nothing.

Difficult feedback to a peer is scored on what they changed

Amazon’s leadership principles describe the trait this question checks. The page carries no date. Under Earn Trust, leaders “speak candidly” and “treat others respectfully.” Have Backbone adds that leaders challenge decisions “even when doing so is uncomfortable or exhausting.”

Both lines describe how you speak. The interviewer still needs evidence that speaking worked. With a direct report, you could point to a review or a plan you set. With a peer, you have no such lever. They changed because they chose to. So the follow-up “what changed?” is where the score is decided.

That’s also why “it went well and we’re still friends” is a weak ending. It describes the relationship. It doesn’t say what the other engineer now does differently.

The release-day pull requests that kept shipping bugs

Here is the example for this post. You’re a mid-level Android engineer on a food delivery app. The team cuts a release every Thursday at 5 p.m. Lena is another mid-level engineer on the same team. Neither of you manages the other.

Lena is fast and well liked. But her pull requests often arrive on Thursday afternoon at 1,000 lines or more. Reviewers are rushed, so they approve after a skim. In one month, two of those pull requests shipped regressions. One crashed checkout when a coupon field was empty. You had approved that one yourself, after about twenty minutes.

Put your own rushed approval in the answer early. It was part of the problem. Admitting it first makes the request you made later easier to trust.

Why the first comment in the pull request thread changed nothing

Your first attempt was a comment on Lena’s next large pull request. You wrote that it was too big to review properly and always came in too late. Other reviewers could see it. Lena replied that product had changed the spec on Wednesday. Then she merged after another engineer approved it.

Google’s code review guide explains part of the failure. The guide carries no date. It advises commenting on the code instead of the developer. It says this matters most when a comment could upset someone. Your comment did the opposite. “Always comes in too late” described Lena’s habit, in public, with no request attached.

In the interview, tell this first attempt plainly. It shows you noticed your method failing. It also explains why the second conversation looked different.

The second conversation named one release and asked for one change

GitLab’s handbook page on guidance on feedback was last updated in October 2025. It recommends the Situation-Behavior-Impact model. Anchor the feedback in a time and place. Describe the behavior someone could observe. Then say what it did to others. The page also warns against absolutes like “you always.”

For the second attempt, you asked Lena for fifteen minutes in private. You named one release. It was the Thursday of the coupon crash. The cart change came in at 4 p.m. with about 1,400 lines. You approved it after twenty minutes. The crash reached users that evening. You said your quick approval was your part of it.

Then you made one request. Could she open a draft pull request by Tuesday, even if the spec was still moving? You also asked what made Thursdays hard. Lena said the product manager often confirmed final copy and edge cases on Wednesday. She didn’t want to send work she would have to redo.

The handbook says to allow the receiver time to reflect. So you didn’t push for an answer in the room. Lena agreed to try draft pull requests for the next month.

Report the partial result, then what you did about the rest

Over the next four releases, three of Lena’s pull requests arrived as drafts by Tuesday. That part changed. Their size mostly didn’t. Lena still preferred to ship a feature in one piece.

Say both halves. An interviewer who hears “everything improved” will usually probe. A partial result with numbers gives them less to doubt.

Then say what you did about the size. You stopped treating it as Lena’s problem alone. You offered to review her drafts in two passes. On Tuesday you checked the data and architecture changes. On Thursday you checked only the final diff. No regression from her pull requests reached users in the following six weeks. You also proposed a team rule that release-day pull requests needed a second reviewer. The team adopted it.

The full answer, as you might say it in the round

A peer on my Android team often opened large pull requests on Thursday afternoon. Our release cut was a few hours later. Reviewers were rushed. In one month, two of those changes shipped regressions. One crashed checkout for users with an empty coupon field. I had approved that one after twenty minutes, so I was part of the problem.

My first attempt didn’t work. I left a comment on her next pull request saying it was too big and always late. Other people could see it. Her reply was that the spec had changed. Then she merged anyway. Looking back, I commented on her habit in public and asked for nothing specific.

So I asked her for fifteen minutes in private. I talked about one release. I brought up the cart change that came in at 4 p.m. with about 1,400 lines. I admitted I approved it too fast. Then I asked for one thing. Could she open a draft by Tuesday, even if the spec was still moving? I asked what made Thursdays hard. Her answer was that product often confirmed edge cases on Wednesday. She agreed to try drafts for a month.

Three of her next four pull requests arrived as drafts by Tuesday. The size mostly stayed the same. So I offered to review in two passes, architecture on Tuesday and the final diff on Thursday. None of her changes caused a regression in the next six weeks. I also proposed a second reviewer for release-day pull requests. The team adopted it.

This answer was written as an example for this post. Lena, the app and the numbers are invented. Use your own peer, your own first attempt and the result you actually saw.

The answer admits the reviewer’s own part before any request. It keeps the failed first try in, so the second try reads as a decision. It ends on two numbers and one action, so the follow-up has already been answered. If the interviewer asks about receiving feedback instead, the order changes. Our post on answering the critical feedback question covers that side.

When the problem isn’t one to raise with a peer first

The advice above fits a habit that slows the team. It doesn’t fit every problem. Suppose a peer logged user data. Or suppose they spoke to a teammate in a way that crossed a line. Then a private chat isn’t the right first step. That goes to your manager, or to whoever owns security or conduct on your team.

If your best story looks like that, don’t force it into this question. Tell it under a question about ethics or escalation instead. Our post on influence without authority when it ends in escalation covers that shape. For this question, pick a story where the other engineer could fix the problem once they heard it. Late pull requests and skipped tests fit. A logged password belongs in the escalation story.

1 thought on “Gave Difficult Feedback to a Peer and Nothing Changed? Tell the Second Attempt”

Leave a Comment