Python asyncio.TaskGroup vs gather: when to use which in production?

The critical difference is structured vs unstructured concurrency. With asyncio.gather, if task 2 raises and you don't use return_exceptions, tasks 1 and 3 are left dangling - they're still running but nobody's awaiting them. TaskGroup fixes this by cancelling all sibling tasks when one fails, which is almost always what you want in production. The one case where gather still wins: fire-and-forget fan-out where you genuinely don't care about failures in individual tasks.