New: Voice AI Playbook, 6 short videos →
THE VOICE AI BUILDER'S PLAYBOOK
Learn to ship voice agents that work
Short, practical videos on debugging, evaluating and testing voice agents with Tuner. One real problem per video.
Start here
Find where the time went.
The first video in the series: break one slow call down by stage and see which part of the voice pipeline caused the delay.
01 · LATENCY
Debugging Voice Agent Latency
Silence on a call makes people repeat themselves, talk over the agent or hang up. This video takes one slow call and debugs it in Tuner, step by step, to find exactly where the time went.
3:53 · Watch on YouTube ↗
All videos
Short videos. One problem each.
Watch them in order, or jump to the problem you have today.
01 · LATENCY
Debugging Voice Agent Latency
Silence on a call makes people repeat themselves, talk over the agent or hang up. This video takes one slow call and debugs it in Tuner, step by step, to find exactly where the time went.
WHAT YOU'LL LEARN
– Break a slow turn down by stage: speech-to-text, end-of-utterance, the LLM, tools and text-to-speech
– Check if it's a pattern with P50 vs. P90, then find the root cause with Diagnose Agent
– Keep it fixed with a red flag and an alert when slow calls pile up
3:53 · Watch on YouTube ↗
02 · TOOL CALLS
Fix Voice Agent Tool Call Failures
A tool call can look healthy in your logs and still fail the caller. We walk through three real calls from a housing repair agent, each failing in a different way, and show how to prove and fix each one.
WHAT YOU'LL LEARN
– Missing tool calls, wrong tool inputs and slow tools
– Tell the agent when to call a tool, and give it today's date and time zone
– Make slow tools asynchronous, with a timeout, one retry and a fallback
5:47 · Watch on YouTube ↗
03 · EVALS & GUARDRAILS
Voice Agent Evals & Guardrails
A completed call isn't always a successful call. Learn how to define what success means for your voice agent and check it on every single call with Tuner.
WHAT YOU'LL LEARN
– Evals that check the transcript, tool calls, system prompt and workflow
– Hallucination checks that show the exact sentence as evidence
– Custom evals in plain English, with alerts when a critical eval fails
2:41 · Watch on YouTube ↗
04 · TURN-TAKING
Voice Agent Interruptions
Your voice agent can't feel when you're about to finish talking. It waits for silence, then guesses: too early and it cuts the caller off, too late and every reply feels slow.
WHAT YOU'LL LEARN
– Turn detection and end-of-utterance latency, shown on two real calls
– Silence-based VAD vs. models that use the transcript or the audio itself
– How to ask questions that don't create pauses
4:15 · Watch on YouTube ↗
05 · TESTING
Automate Voice Agent Testing
Real callers hesitate, interrupt and change their minds, and you can't cover every conversation by calling your agent yourself. See how to automate testing with call simulations in Tuner.
WHAT YOU'LL LEARN
– Set up a run with Routine Callers and Pressure Tests
– Read the results and open a failed call to find the cause
– Rerun a focused simulation after the fix to confirm it's gone
5:09 · Watch on YouTube ↗
06 · OBSERVABILITY
Voice Agent Observability
OpenTelemetry, Langfuse and your provider logs cover only one of the four layers you need to monitor a voice agent. This video walks through all four and what each should tell you.
WHAT YOU'LL LEARN
– Technical and business observability: how components perform, and whether the agent did its job
– Data and security controls: PII redaction, retention and hosting
– Proactive testing, dashboards and alerts
4:06 · Watch on YouTube ↗