Why Your Agent Loops
Infinite loops are the most common failure mode in AI agents. Learn the four root causes — ambiguous tools, context pollution, missing stop conditions, and reasoning spirals — with concrete detection and prevention strategies.
Quick Reference
- →Agent loops happen when the model calls the same tools repeatedly without making progress
- →Root cause 1: ambiguous tool descriptions — the model cannot distinguish which tool to use
- →Root cause 2: context pollution — accumulated tool results confuse the model's reasoning
- →Root cause 3: missing or unclear stopping conditions in the system prompt
- →Root cause 4: reasoning spirals — the model talks itself into circles trying to be thorough
- →Always add circuit breakers: max iterations, max tokens, max tool calls, and timeout limits
Recognizing Agent Loops
An agent loop looks like this: the user asks a question, the agent calls a search tool, gets results, decides the results are not good enough, calls the search tool again with a slightly different query, gets similar results, and repeats 15 times before hitting a token limit or timeout. The user sees a spinning indicator for 60 seconds and gets a mediocre response. This is the most common agent failure in production.
Not all repeated tool calls are loops. An agent that searches, reads a document, searches again based on what it learned, and then answers is doing legitimate iterative research. The key difference: in a productive iteration, the context changes meaningfully between calls. In a loop, the agent is repeating the same actions with trivially different inputs.