Blog

The voice AI playbook.

Field notes for the leaders, operators, and engineers driving AI transformation — what actually moves the number when voice AI hits production, what breaks at scale, and what your peers are shipping this quarter.

Why sub-400ms latency is the difference between a deal and a hangup

Three architectural choices that cut 240ms out of our voice pipeline — and why every millisecond past 400 quietly costs you bookings.

Why one voice model can't run every call type

How we route TTS by call type — and why a one-size-fits-all voice quietly tanks conversion on the calls that matter.

The HIPAA voice AI checklist your security review will ask for

Everything we built and audited before going live in healthcare — the controls your CISO and the BAA will both want documented.

What an AI receptionist rollout actually looks like, week by week

Day 0 scope, Week 1 integrations, Week 2 script tuning, go-live — what a real voice AI deployment costs in calendar time and team hours.

Function calling vs. tool calling — and why the choice changes your CRM integration cost

When to use each, with code samples — plus the architectural decision that determines how much it costs to wire up the next system.

Voice AI's biggest unsolved problem: barge-in

Why interruption handling is harder than turn-taking — and why the agents that get it wrong feel robotic no matter how fast they respond.

Scaling to 40,000 concurrent calls — and what broke first

The architectural failure modes that surfaced when we 10x'd peak load — and what to ask any voice AI vendor about their ceiling.

The voice AI rollout playbook for insurance carriers

Speed-to-lead, renewals, FNOL — what to deploy first, what to expect from each, and how the unit economics actually pencil out.

Voice AI latency benchmarked — harmony, Bland, Retell, Vapi on the same workload

Methodology, raw numbers, reproducible scripts — the data your evaluation team can rerun themselves, with no vendor spin.

Subscribe

One post a week. No spam.

See voice AI in production — this week.

Pick a use case, see the demo, run it in your environment. No telephony glue required.

Book a demo