Why did my AI agent get slower even though nothing is throwing errors?

Usually one of three things: the model behind the call got slower on the provider’s own schedule, a tool the agent calls started dragging under its own load, or the agent’s prompt grew. None of the three throws an exception, which is why nothing in your stack flagged it.

A slowdown and a failure are different events, and most monitoring is built to catch the second. The agent still returns a completed response, so no test breaks; the request just takes longer than it used to.

The prompt case is the easiest one to miss, because nobody ships it deliberately. More retrieved context, more conversation history carried forward, and every added token costs processing time on top of whatever the model was already taking.

None of that shows up in an error rate. It only shows up when you compare how long things take now against how long they used to take for the same kind of request.

keep reading

More on this.

Can duration_drift catch a slowdown that happens gradually, not all at once?Yes. It compares against a rolling recent baseline that catches sudden breaks and one pinned at the last deploy that catches the creep the rolling one absorbs.answer →Can duration_drift tell me which tool call made a turn slow?Yes. It watches turn duration and per-tool duration together, so a slow turn and the specific tool call behind it land in the same finding.answer →Does duration_drift use a fixed latency threshold?No. There's no absolute cutoff; each call site and tool is compared only against its own recent history, since normal speed varies enormously by workload.answer →Everything on tessary duration driftThe definition, and every question answered under tessary duration drift.concept →

Send us the traces you already emit.

book a demo