A coding agent turns one task into hundreds of model calls. Here is how a gateway gives each run a hard ceiling that holds under burst, a fallback path that does not strand a half-finished edit, and a cost you can read per task.
An agent sends hundreds of calls without a human in the loop, so the loop bug you have not written yet is a billing event. Give each agent role its own key with its own rate ceiling and spend cap, enforced at the gateway rather than in the agent code that has the bug.
One user action becomes fifty model calls across four agent roles. Here is how to attribute that spend by role, by run and by step, reconcile it against the ledger, and put a ceiling under it that application code cannot bypass.
An agent run is not one request, it is eleven. Here is how to route each step to the model that fits it, keep the run alive when a provider fails mid-chain, and read one settled cost for the whole run instead of guessing.
