Our agent sometimes burns 20 tool calls on flaky APIs. Hard caps feel arbitrary. Looking for practical budgets (tokens, wall time, tool diversity) that still allow recovery.
How do you bound agent tool loops without killing useful retries?
by Dev Patel
Answers (1)
Budget by ‘unique tool + args fingerprint’ plus wall clock. Identical retries should escalate once, then fail closed. Log the loop reason so product can tune caps per workflow.