Field notes

Field notes on Voice AI.

Practical writing on testing, monitoring and improving voice agents — from the team building the platform.

Testing endpointing and turn detection in voice AI agents
Voice AI Testing

Testing endpointing and turn detection in voice AI agents

Endpointing is where voice agent latency and false interruptions hide. Here's how to test turn detection across VAD, STT, and semantic models before you ship.

Read the piece· 10 min read
Vendor-run simulations aren't your acceptance test
Evaluation Strategy

Vendor-run simulations aren't your acceptance test

OpenAI Presence bundles simulations, graders, and guardrails into a managed voice agent. Here's the independent acceptance test enterprise buyers still owe themselves.

Testing warm transfer and human handoff in voice AI agents
Voice AI Testing

Testing warm transfer and human handoff in voice AI agents

Warm transfer is where voice agents lose the caller. Here's a technical test plan for triggers, context capture, whisper briefings, and post-bridge behavior.

Codex proposes, callers dispose: testing AI-authored voice agent updates
Evaluation Strategy

Codex proposes, callers dispose: testing AI-authored voice agent updates

OpenAI Presence turned "AI proposes an agent update, humans approve it" into a shipping pattern. Simulation testing is what makes the review step real.

Testing voice agents against background noise
Voice AI Testing

Testing voice agents against background noise

Real callers dial from cars, cafes, and warehouses. Here's how to build an SNR-tiered noise test suite for voice agents, and how to run it before every deploy.

Sample-based QA doesn't work for voice AI agents
Evaluation Strategy

Sample-based QA doesn't work for voice AI agents

Contact centers sampled 1-3% of calls for decades. Voice AI agents change too fast for that to work. Here's the two-part QA pattern that does.

This is Roark v2
Product

This is Roark v2

Roark v2 is a ground-up rebuild of our voice-AI testing platform: simulation templates, a new look with dark mode, faster reporting, and self-serve pay-as-you-go pricing.

Testing tool calls in voice agents
Voice AI Testing

Testing tool calls in voice agents

Voice agents now book, refund, and update records mid-call. Here's how to test tool calls (functions, MCPs) end to end before your agent invents an argument on air.

Your voice agent's AI disclosure is a regression test, not a prompt line
Voice AI Testing

Your voice agent's AI disclosure is a regression test, not a prompt line

State disclosure laws now vary by jurisdiction, trigger, and timing. Here's how to regression-test AI voice agent disclosure logic before a prompt tweak breaks it.

The two-minute voice agent
Voice AI Testing

The two-minute voice agent

Grok Voice Agent Builder, Retell Conductor, and GPT-Live-1 all landed in two weeks. Building a voice agent takes minutes now. Testing it doesn't.

Testing full-duplex voice agents after GPT-Live
Voice AI Testing

Testing full-duplex voice agents after GPT-Live

GPT-Live made full-duplex the baseline for voice AI. Here is how testing has to change when your agent listens and speaks at the same time.

Voice agents now ship in two minutes. Testing is the new bar for launch.
Voice AI Testing

Voice agents now ship in two minutes. Testing is the new bar for launch.

No-code builders from xAI, Retell, and OpenAI cut voice-agent build time to minutes. Here is the pre-launch testing bar that separates demos from production.

Testing full-duplex voice agents: what changes when your model listens and speaks at once
Voice AI Testing

Testing full-duplex voice agents: what changes when your model listens and speaks at once

GPT-Live made full-duplex the default. Here's what breaks in your voice-agent eval stack, and what audio-native testing has to look like now.

AI disclosure is a testable requirement. Here's how to prove your voice agent gets it right.
Compliance & QA

AI disclosure is a testable requirement. Here's how to prove your voice agent gets it right.

State-by-state AI voice disclosure laws are stacking up fast. Here's how QA and product teams turn disclosure into a testable, provable requirement on every call.

Roark vs Hamming: which voice AI testing platform fits your team?
Platform Comparison

Roark vs Hamming: which voice AI testing platform fits your team?

An honest comparison of Roark and Hamming for voice agent testing — where each concentrates on simulation, audio-native evaluation, replay, integrations, and compliance.

Roark vs Cekura: which voice AI testing platform fits your team?
Platform Comparison

Roark vs Cekura: which voice AI testing platform fits your team?

An honest comparison of Roark and Cekura (formerly Vocera) for voice agent testing — simulation depth, audio-native metrics, production replay, integrations, and compliance.

Roark vs BlueJay: correcting the record
Platform Comparison

Roark vs BlueJay: correcting the record

A widely shared "BlueJay vs Roark" pricing breakdown gets Roark wrong — the pricing, the tiers, the metrics, and the compliance. Here is each claim and what is actually true.

Testing voice agent pronunciation of medication names
Voice AI Testing

Testing voice agent pronunciation of medication names

How to build a regression test harness for medication name pronunciation on a voice agent: golden sets, LASA adversarial pairs, dictionaries, and scoring on real audio.

The voice agent launch-readiness checklist
Evaluation Strategy

The voice agent launch-readiness checklist

A PM's launch-readiness checklist for voice AI agents: five risk domains, the tests that catch each one, and what "careful adoption" actually means.

How to test voice agent barge-in and interruption handling
Voice AI Testing

How to test voice agent barge-in and interruption handling

Barge-in is where voice agents feel human or robotic. Here is how to test interruption handling systematically instead of hoping it works in production.

How to test Vapi voice agents
Voice AI Testing

How to test Vapi voice agents

Why Vapi agents break in production, what Test Suites, Evals, and Simulations actually cover, and how to test the full audio path with real phone calls.

How to test Retell AI voice agents
Voice AI Testing

How to test Retell AI voice agents

Where Retell agents fail, what Retell's native Playground, simulation, and batch-test API cover, and how to run phone-level voice simulations and regression tests.

Roark vs Coval: which voice AI testing platform fits your team?
Platform Comparison

Roark vs Coval: which voice AI testing platform fits your team?

An honest, source-linked comparison of Roark and Coval for voice AI testing — how each platform simulates and evaluates voice agents, and where Roark goes deeper on real-call simulation and audio-native scoring.

Mastering AI test agents in Voice AI testing
Voice AI Testing

Mastering AI test agents in Voice AI testing

How AI test agents change Voice AI testing — autonomous simulation, scalable coverage, and audio-native quality metrics.

The top 5 Voice AI agent platforms in 2025
Platform Comparison

The top 5 Voice AI agent platforms in 2025

A strategic comparison of Leaping AI, Bland AI, Retell AI, Parloa and Vapi — and how to pick the right one.