Phone agents change first touch. They do not change the offer.

Phone agents change who dials. They do not change the offer.

TechCrunch reported on Sep 24, 2026 that Google is testing Gemini "Call for Me" on Pixel. The agent dials businesses from your number, navigates menus, waits on hold, keeps a live transcript, and lets you take over. That is a different receipt from yesterday's ChatGPT mobile voice story. This is outbound phone dialing on behalf of the buyer, not a voice chat inside an app.

Phone agents changed who dials. They did not change the offer. Disclosure plus purchase label still own the meeting.

How is this different from mobile voice agents?

Voice on mobile changes the channel inside a product. Call for Me changes the call itself.

Your number rings someone else's desk. The agent speaks for a human. Trust is on the line before the offer is. If the agent sounds slick and the ask is vague, you burned a live human's attention with theater.

B2B GTM will copy this pattern fast: SDR agents that dial, AE agents that "confirm," founder agents that "follow up." Same Convert problem as cold email. New surface. Same need for a clear purchase label.

What stays the same in Convert?

Desire. Mechanism. Ask.

The buyer still wants the outcome under the purchase label, not a tour of your voice stack. The grunt test still applies: what do I get, how does life improve, what pain leaves, why trust you. If the person who picks up cannot grunt that in one breath after the agent opens, the dial did not save you.

I have watched teams swap email for dialers and keep the same "we help GTM teams with AI" open. Channel changed. Note still could have gone to anyone. Cold email that books a meeting wins for the same reason a phone agent will win: signal match, plain outcome, credible next step.

Why do agent dials spoil trust?

Because a silent bot on a human's number feels like a trick.

Google's own framing includes live transcript and takeover. That is the minimum pattern. Same law as human disclosure for GTM agents: say when an agent is speaking, and give a path to a human. Buyers punish feeling cheated more than they punish knowing an agent dialed.

Forrester-style ~84-day B2B cycles already punish soft opens. An agent that talks for three minutes without naming the signal, the outcome, or the ask adds cost to a cycle that was already long.

Salesforce still puts about 27% of reps at quota. An agent dial that books a calendar and spoils trust on the other nine accounts is not "AI GTM." It is a domain-burn problem with a prettier voice.

How should first touch work on a phone agent?

1. Open with disclosure. "This is an assistant calling for [Name] about [specific signal]." No fake human.
2. Name the signal. Website visit, G2 category look, competitor thread. If you cannot name it, do not dial.
3. State the purchase label and the ask. One outcome. One next step. Book fifteen minutes or get off the line.
4. Keep takeover live. Human can grab the call when judgment shows up. Untested takeover is theater.
5. Log the transcript against the CRM row. If the agent lied or drifted, you need the receipt to fix the rails.

How do I pressure-test this week?

Script five agent dials the way you would script five cold emails.

Ask: is disclosure first? Is the signal named? Does the purchase label show up in the first thirty seconds? Can a human take over without restarting trust? Would you accept that open if it hit your own cell?

If three of five fail, freeze agent dials. Keep humans on the phone. Fix the message. Then reopen dials one segment at a time.

Would you rather an agent that discloses, names the signal, and books a clear meeting, or a perfect TTS voice that burns the account before a human ever speaks?

Adapt or fail. Who dials changed. The offer did not. Convert still owns the meeting.

Start Signals, Convert, Grow

FAQ

Is Call for Me a B2B sales product?

Not yet. It is a consumer-facing business dial experiment. The GTM lesson is the pattern: agent dials from a human number with transcript and takeover. Copy the rails, not the demo.

Should my SDR agent pretend to be human?

No. Disclosure first. Pretending is a trust tax you pay later in the cycle.

Does a phone agent replace cold email?

No. It is another vehicle. Signal match and purchase label still decide. Pick the vehicle after the message is clear.

How does this fit Signals, Convert, Grow?

Signals decides who is worth a dial. Convert decides what the agent may say and ask. Grow decides when volume scales. Skip Convert and you scale spoiled trust.