THE EDITOR'S DESK

A closer
listen.

The voice AI stories
worth following.

Ideas, observations and practical questions about the conversations shaping business.

FIELD NOTES / VOL. 0103 STORIES · ONE CLOSER LOOK

The New Wave of AI Voice Startups

Smaller companies are creating specialized solutions for real-world conversations.

Start with the job, then the technology.

A restaurant taking an order, a contractor capturing a new inquiry, and an online store answering an order-status question have different conversations. A useful discovery process starts by naming that conversation.

Look for focus.

Our watchlist groups companies by the workflows they describe: everyday phone answering, home-service inquiries, restaurant calls, e-commerce support, and custom business workflows. These are starting points for exploration, not measured endorsements.

Three questions to bring to a demo

  1. Can you show the full workflow, including what happens in my business systems after the call?
  2. What happens when the caller changes their mind, interrupts, or needs a person?
  3. Who configures, monitors, and improves the agent after launch?

Keep a shortlist based on the job you need done. Then test the same scenario with every provider.

This is a discovery guide. Independent testing and performance rankings for the 2026 watchlist are pending. See the profiles.

HEARDFIRST / THE EDITOR'S DESKNext story

What Makes an AI Voice Agent Actually Good?

Why latency, interruption handling and task completion matter more than a polished demo.

A voice is only part of the experience.

Listen to how the agent moves through a conversation. Can a caller interrupt? Does the agent remember a corrected name or date? Does it explain when it cannot complete a request?

Watch the outcome.

A confident confirmation is not enough. In a test environment, verify that the appointment, inquiry, or handoff actually reached the intended destination with the correct details.

Keep the conditions consistent.

Use the same script, the same background noise, and the same interruption point for each agent. Record the configuration and date alongside observations. A single test call provides an initial snapshot, not proof of reliability across all callers or situations.

Our five planned checks

  • First-word latency
  • Barge-in performance
  • Conversation recovery
  • Task completion
  • Audio quality

Measured results will be added to the Rankings after testing.

HEARDFIRST / THE EDITOR'S DESKNext story

The Real Cost of Voice AI

Understanding setup fees, usage charges, integrations and ongoing support.

Ask for a complete cost breakdown.

A headline subscription price is only one line of a proposal. Ask the provider to identify which of the following items are included, separately billed, or not applicable.

  • Setup: workflow design, agent configuration, testing, and launch support.
  • Recurring platform fees: the service, number of agents, and included support.
  • Usage: how calls, minutes, messages, or conversations are counted and billed.
  • Connections: phone services, business-system integrations, and maintenance.
  • Ongoing changes: new workflows, revisions, monitoring, and quality review.

Compare the same scenario.

Give every vendor the same expected call volume, typical call length, required integrations, and support needs. Request a written estimate for that scenario and a second estimate for a busier month.

Ask what happens at the limits.

Confirm overage rates, minimum commitments, cancellation terms, number ownership, and how you can export your data. If a fee is unclear, ask for the calculation in writing.

Published starting prices for each watchlist company are in the profiles. Current fees and contract terms must be confirmed directly with each provider.

HEARDFIRST / THE EDITOR'S DESKBack to the top
FROM READING TO LISTENING

Find your next
conversation.

Meet the companies on our radar and explore the work behind their voices.

Explore the Companies ↗
HEARDFIRST / DISCOVERY NOTES