TL;DR: Pick a real, substantive technical disagreement. Show that you made your case with data and a proposed test rather than seniority or opinion, that you genuinely listened to the other side, and that once a decision was made you committed fully even if it was not your pick. Use STAR, say "I," and end with the outcome and what you learned.
How to approach it
Choose a disagreement where you were partly wrong or the outcome was nuanced; an "I was right and they were idiots" story reads as a red flag. Structure it STAR and spend the time on how you resolved it, because the method is the competency being scored: conflict resolution and influence without authority.
A strong answer
The arc interviewers reward:
- Situation and stakes. A concrete technical fork with real consequences: "We disagreed on whether to fine-tune or ship a RAG approach against a customer deadline." Make clear it mattered.
- Your position, held with humility. State what you believed and why, then immediately show you sought to be right, not to win: "I could have been wrong about the eval cost, so I proposed we settle it with data rather than argue."
- Resolution by evidence. "I built a small benchmark on a representative slice, we agreed on the metric beforehand, and the numbers showed RAG hit the quality bar faster." Resolving disputes with a quick, pre-agreed test rather than status is the single strongest signal, and it is exactly what labs and big tech look for.
- Genuine listening. Show the other person had a real point you incorporated: "Their concern about latency was valid, so we added a caching layer."
- Disagree and commit. If the decision went against you: "Leadership chose the other path; I committed fully, owned the execution, and it worked." Backbone to voice it, maturity to commit.
- Outcome and lesson. Quantify the result and name what you would do earlier next time (often "align on the deciding metric sooner").
That arc shows you can push hard on substance, change your mind on evidence, and still row in one direction once a call is made.
Key takeaways
- Win on a pre-agreed metric and a small benchmark, never on volume or seniority.
- Show you absorbed at least one valid point from the other side; pure victory stories read as low collaboration.
- If overruled, commit visibly and own execution; the panel is scoring disagree-and-commit, not who was right.
- Close with a quantified outcome and a specific lesson.
What interviewers probe next
- "What if you had been overruled and it failed?" Show you would surface the risk early, commit anyway, and run a blameless retro, not say "I told you so."
- "How do you disagree with someone much more senior?" Privately first, with data and framed as a shared goal; respect the decision rights while still voicing the concern.
- (Amazon) "Which Leadership Principle is this?" "Have Backbone; Disagree and Commit," and often "Are Right, A Lot" and "Earn Trust."
- "How did the relationship hold up?" Strong answers show trust intact or strengthened afterward.
Common mistakes
- A story where you were simply right and the other person foolish; reads as low collaboration.
- Winning by seniority or volume instead of evidence.
- "Disagree and commit" stories where you actually disengaged or quietly undermined the decision.
- No "I" ownership, or no outcome and lesson, leaving it as a vague anecdote.