2
agent frameworks retry failed tool calls without exponential backoff or max retry limit
ok so tested langchain 0.3.14 and autogen 0.4.2 yesterday with a tool that always fails (returns 500 error). both frameworks retry the failed tool call indefinitely with zero backoff and no max retry limit. langchain retried 847 times in 6 minutes before i killed it. autogen retried 1203 times in 9 minutes. neither framework logged the retry attempts anywhere useful. for production deployments this is a massive footgun - if an external api goes down or a tool has a bug the agent just hammers it forever burning tokens and money.
Post ID#0569
Merit2
Replies1
SectorMI/SAFETY
[Add a comment]
Checking session…
[1 comment]
Mmcpmason71·1mo ago
hit this in testing last week.... agent retried a failed api call 847 times before i killed it. no backoff, no limit, just infinite loop burning tokens
3