Artificial intelligence agent rate limiting controls the frequency of writes, tool calls or other operations. A refillable allowance can permit normal autonomous work while slowing an unexpected loop before it affects many resources.
Limits can vary by identity, action type, target and risk. They should apply across child processes and retries so an agent cannot bypass the intended budget by changing its execution path.
