How to Fix 'agent infinite loop' in AutoGen (Python)

What the error means

agent infinite loop in AutoGen usually means your agent conversation never reaches a stopping condition, so the framework keeps routing messages forever. In practice, this shows up when AssistantAgent, UserProxyAgent, or a custom agent keeps replying without a termination signal, especially in initiate_chat() or group chat flows.

You’ll often see this after a few back-and-forth turns, then the run never ends until you kill the process or hit a max-iteration guard.

The Most Common Cause

The #1 cause is a missing or broken termination condition. In AutoGen, if your assistant keeps generating a response and your user proxy keeps auto-replying, you have created an endless loop.

Here’s the wrong pattern and the right pattern side by side:

# WRONG: no reliable stop condition
from autogen import AssistantAgent, UserProxyAgent

assistant = AssistantAgent(
    name="assistant",
    llm_config={"config_list": [{"model": "gpt-4o-mini"}]},
)

user_proxy = UserProxyAgent(
    name="user_proxy",
    human_input_mode="NEVER",
    code_execution_config=False,
)

user_proxy.initiate_chat(
    assistant,
    message="Summarize this policy document."
)

# RIGHT: explicit termination handling
from autogen import AssistantAgent, UserProxyAgent

def is_termination_msg(msg):
    content = msg.get("content", "")
    return isinstance(content, str) and content.strip().endswith("TERMINATE")

assistant = AssistantAgent(
    name="assistant",
    llm_config={"config_list": [{"model": "gpt-4o-mini"}]},
)

user_proxy = UserProxyAgent(
    name="user_proxy",
    human_input_mode="NEVER",
    max_consecutive_auto_reply=5,
    is_termination_msg=is_termination_msg,
    code_execution_config=False,
)

user_proxy.initiate_chat(
    assistant,
    message="Summarize this policy document. End with TERMINATE."
)

If you’re using AutoGen’s standard chat loop, the assistant must know how to end. If you’re using a tool-calling setup, make sure tool results don’t keep re-triggering the same prompt without ever producing a terminal answer.

Other Possible Causes

1) Your system prompt encourages endless refinement

A common trap is asking the model to “keep improving until perfect” with no exit rule.

llm_config = {
    "config_list": [{"model": "gpt-4o-mini"}],
}

assistant = AssistantAgent(
    name="assistant",
    system_message=(
        "Keep revising the answer until it is perfect. "
        "Do not stop unless you are fully satisfied."
    ),
    llm_config=llm_config,
)

Fix it by adding an explicit completion rule:

assistant = AssistantAgent(
    name="assistant",
    system_message=(
        "Answer once. If complete, end with TERMINATE. "
        "Do not continue refining after that."
    ),
    llm_config=llm_config,
)

2) GroupChat speaker selection keeps selecting the same agent

In GroupChat, a bad speaker_selection_method can bounce messages between agents forever.

from autogen import GroupChat, GroupChatManager

groupchat = GroupChat(
    agents=[agent_a, agent_b],
    messages=[],
    speaker_selection_method="round_robin",  # can loop if neither agent terminates
)

manager = GroupChatManager(groupchat=groupchat, llm_config=llm_config)

Fix by adding termination checks and limiting rounds:

groupchat = GroupChat(
    agents=[agent_a, agent_b],
    messages=[],
    max_round=6,
)

manager = GroupChatManager(groupchat=groupchat, llm_config=llm_config)

3) Tool execution returns output that re-triggers the same request

This happens when an agent calls a tool, receives output, then asks for the same tool again because your prompt doesn’t distinguish “done” from “need more data”.

assistant = AssistantAgent(
    name="assistant",
    llm_config=llm_config,
)

user_proxy = UserProxyAgent(
    name="user_proxy",
    human_input_mode="NEVER",
    code_execution_config={"work_dir": "coding"},
)

A better pattern is to constrain tool usage and define completion clearly:

assistant = AssistantAgent(
    name="assistant",
    system_message="Use tools only once per task. Return final answer after tool output.",
    llm_config=llm_config,
)

4) Your custom reply function always returns a response

If you override reply logic and never return None, AutoGen assumes there is always another message to send.

def reply_func(recipient, messages=None, sender=None, config=None):
    return {"content": "Working on it..."}  # never stops

Fix it by returning None when done:

def reply_func(recipient, messages=None, sender=None, config=None):
    last = messages[-1]["content"]
    if "TERMINATE" in last:
        return None
    return {"content": "Final answer. TERMINATE"}

How to Debug It

1. •

   Check whether termination is actually reachable

   • •Search for is_termination_msg, "TERMINATE", or equivalent stop logic.
   • •If you don’t have one, that’s likely your bug.
2. •

   Print every turn

   • •Log sender name, recipient name, and message content.
   • •You want to see whether the same two agents are repeating the same intent.
3. •

   Reduce to two agents

   • •Remove tools, group chat managers, retrieval layers, and custom reply hooks.
   • •If the loop disappears in a minimal setup, add pieces back one at a time.
4. •

   Inspect auto-reply settings

   • •Look at human_input_mode, max_consecutive_auto_reply, and any custom reply_func.
   • •A common failure mode is human_input_mode="NEVER" plus unlimited auto responses.

Prevention

• •Always define an explicit end condition:
  • •Use "TERMINATE" or a structured final message.
• •Put hard limits on conversation length:
  • •Set max_consecutive_auto_reply and max_round.
• •Make prompts state completion rules:
  • •“Answer once and stop” beats “keep iterating until perfect.”

If you want production-safe AutoGen flows in Python, treat termination as part of your protocol design. The model will not guess when your workflow is done unless you make that rule unambiguous.

Keep learning

• •The complete AI Agents Roadmap — my full 8-step breakdown
• •Free: The AI Agent Starter Kit — PDF checklist + starter code
• •Work with me — I build AI for banks and insurance companies

By Cyprian Aarons, AI Consultant at Topiax.