Tell me about a disagreement with a coworker where you were the one who changed their mind.

A developer on a partner squad, no shared manager and not my codebase, kept skipping the contract checks we had put around a shared policy-pricing service. When they went red he marked them pending and promised to fix them later, and later kept not arriving. Telling him he was doing it wrong was never going to work, and I had no standing to insist. So I asked to sit in on one of his sprint reviews, and that is where the actual reason showed up: the contract pack took eleven minutes locally and he was cutting three or four small releases a week. He was not careless, he was rationally impatient. I spent an afternoon splitting the pack so the pricing contracts alone ran in under ninety seconds, and I brought him that with a single ask: run just that slice before you push, for a fortnight, and tell me if it gets annoying. That was a cheap yes, and it cost him nothing to abandon. Over that quarter, defects escaping from that service into the shared integration environment fell from eleven to three. He started writing contract cases himself, and when a new engineer joined his squad he was the one who explained why they mattered, which is a better outcome than me having been right.

As test architect on an enterprise claims platform, I had three squads with genuinely different beliefs about what a regression suite was for. The person I most needed to move was the strongest engineer on the busiest squad, who believed a careful manual pass before release was safer than anything automated. He did not report to me and did not need my agreement for anything. I did not open with a standard, because a standard from someone with no authority is just an opinion with formatting. I asked him to sit down with me and read every defect that had reached the production support queue over the previous half-year, thirty-four of them. Twenty-one were regressions in paths his own exploratory passes had covered two or three releases earlier and stopped covering, simply because there is more surface than any person can hold in their head. That reframed the argument from automation versus craft into coverage decay, which was already his language. Then I asked for something small and flattering instead of a mandate: would he own the list of paths that must never regress? He wrote it. Twenty-nine paths, his judgement, our pipeline. Over the following two halves the support queue took twelve escaped defects rather than thirty-four, and he sold the approach to the other squads far better than I could have. Hand the sceptic the pen has been my default move since.