# how do you decide when a voice agent should just hang up? — retained record

> Retained reddit record from reddit:AI_Agents.

Canonical: https://highsignal.app/data/records/c79b5ce6ffcaaf93

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:AI_Agents)  published 2026-10-21 14:54 ingested 2026-10-22 02:36

[original source ↗](https://www.reddit.com/r/AI_Agents/comments/1xbpcs4/how_do_you_decide_when_a_voice_agent_should_just/)

## retained text



```
I keep going back and forth on this! Want to hear how other people handle it. When I build phone agents, the hardest edge case isn't the happy path. It's knowing when the call should end.

  
If the caller goes quiet, do you wait. For how long before you treat silence as done. I've tried a few heuristics based on turn count and VAD timing; they all feel a bit arbitrary.

  
The failure mode I worry about most is the opposite of a bad hangup. An agent that keeps talking past the point where a human would've ended things. That reads as broken even if each response was fine on its own. It's a pacing problem, not a logic problem. I don't think STT accuracy or TTS quality fixes it on its own.

  
Do you use a fixed number of silent turns as a cutoff. Do you have the agent ask an explicit are you still there before ending. Does this change if the call came in over SIP versus a WebSocket setup. I'm curious whether people treat this as prompt engineering or something handled in the call orchestration layer, outside the model entirely.

  
Has anyone actually tested different silence thresholds end to end. Three seconds versus ten feels like it should matter a lot! Would love to hear what's worked, especially past a prototype.
```



document fetched 2026-10-22 00:19 · canonical https://reddit.com/r/AI_Agents/comments/1xbpcs4/how_do_you_decide_when_a_voice_agent_should_just
