What is Exponential backoff?
Exponential backoff is a retry strategy that waits progressively longer between attempts, doubling the delay each time, usually with random jitter, so a client recovers from transient failures without overwhelming an already-struggling service.
Exponential backoff is how a well-behaved client retries a failed request. Instead of hammering a service the instant a call fails, it waits a short delay, then doubles that delay on each subsequent retry, 1s, 2s, 4s, 8s, up to a cap, giving an overloaded or rate-limited dependency time to recover. The key refinement is jitter: adding randomness to each delay so that many clients that failed at the same moment do not all retry in lockstep and create a synchronized thundering herd that knocks the service down again. Backoff is the standard response to transient errors, network blips, timeouts, HTTP 429 rate-limit and 503 unavailable responses, but it should be applied only to failures that are actually retryable and ideally only to idempotent operations, so a retried write does not duplicate an effect. It is everywhere in agent and MCP systems because LLM and tool APIs frequently rate-limit: SDKs commonly retry with exponential backoff and jitter by default, and pairing retries with an idempotency key makes them safe. Combined with timeouts and a circuit breaker, backoff is a core piece of building resilient clients.