What is Rate Limiting?

Every serious API enforces limits, usually as a quota per window: requests per minute, per method, per token. Exceed it and you get HTTP 429, often with a Retry-After header telling you how long to wait. The correct response is exponential backoff with jitter — wait a bit, retry, wait longer, and randomise so a fleet of clients doesn't all retry in lockstep. Hammering a 429 is how a temporary throttle becomes a ban. Limits also come as concurrency caps and daily quotas, not just per-second bursts.

Agents hit rate limits harder than humans do, because a loop that lists a mailbox page by page can issue a hundred calls in the time a person makes one. Connectors are the right place to handle this: they own the retry, the backoff and the queueing, so an agent sees a slower tool rather than a failure it has to reason about. Two things to watch — retries and idempotency are linked, since a retried write can duplicate; and backoff is a direct contributor to a run's end-to-end latency.