I called code assistants a gimmick until one caught a bug that cost my team 2 days in March
Back in January I told my lead that AI code tools were just autocomplete with good marketing, and I refused to turn one on for our payment service repo. Then in March a teammate ran a scanner over a refund handler and it flagged a race condition on a retry path that had already burned us for 2 days of debugging, and it pointed to line 212 exactly. I spent an afternoon checking that it was not a false alarm, and sure enough, two threads could both read the same stale order state in a narrow window under load. I still review every suggestion line by line, but I was the one holding up the team. How do you all decide which of these suggestions are worth trusting in a payment flow?