How phone voice agents work, and how they fail

Where to start: what to check and how a test works.

When the agent speaks, when it stops, and the wait the caller hears.

Narrow lines, regional accents and background noise.

What to check on a named platform.

Latency, interruptions, control, cost, and what each design means for testing. 7 minute read.

Wrong passes, partial answers, persuasive callers and spelled letters, with a test list. 7 minute read.

Barge-in, backchannels, false stops, echo, and what to measure. 7 minute read.

Evals, simulations, call logs, webhooks and the settings to check on a real call. 7 minute read.

Simulation, web and phone tests, call history and post-call analysis. 8 minute read.

Simulation tests, evaluation criteria and post-call webhooks. 6 minute read.

Unit tests with a judge, simulations, metrics and real SIP calls. 7 minute read.

Evals, metrics, tracing and real calls over a phone transport. 7 minute read.

Six options, who each suits, and what to test before you switch. 6 minute read.

Build, control, telephony, latency claims, pricing and testing, side by side. 6 minute read.

How Vapi, Retell and Bland bill a call, with a worked 3-minute example. 6 minute read.

Manual, scripted and simulated callers, what to check, and how many calls to make. 6 minute read.

The benchmark method, the test conditions and how to rerun a result.

A worked example with five made-up agents and planted faults. Example data, no run yet.

Speed, turn-taking, hearing, staying on task, handover and recovery, each with a test. 6 minute read.

When typed tests are enough, which faults only show up in audio, and how to mix them. 7 minute read.

What a good handover to a person looks like, and how agents get stuck in loops. 7 minute read.

An accent mix from real callers, noise at set levels, phone audio and fair reporting. 7 minute read.

What a trace and a span are, and how to use them when you test. 7 minute read.

Thirty terms you will meet when building or testing a phone voice agent.