The problem

Monday, 9:02. A customer deploys a sync script with a bug: when a call fails, it retries at once, forever. Within a minute it sends 40 000 requests to your public API.

Your 8 app servers spend their time on that one API key. Requests from the 3 999 other customers start to time out, and your support inbox fills up.

The contract says each key may make 100 requests per minute. Nothing in the code checks it yet.