hard20 pts · first answer countsagentsreliability

Agent stuck in a loop

An agent sometimes calls the same search tool again and again, burning thousands of tokens without finishing. What is the best set of fixes?

A.Remove the search tool entirelyB.Raise the temperature so it tries different thingsC.Increase the context window so the loop can continue longerD.Set step and token budgets, detect repeated identical calls, return clear tool errors or results, and fall back to a safe answer or a human when limits are hit

← Shipping a prompt change safely Upgrading the embedding model →