Instinct vs Meta Muse: AI Assistants Can Now Call You

Update, Sept. 17: Meta's calling test is more locked down than the announcement suggested: a private test of Muse found it limited to verified US business numbers — it refused requests to call a personal number and declined a one-time demo call without a confirmed eligible number, and its terms bar bulk calls without the consent required under the Telephone Consumer Protection Act (RuntimeWire). Separately, the economics behind Instinct's push are sharpening: Instinct raised $350 million in August at a $2.5 billion valuation and is now reportedly in talks for about $1 billion at a $10 billion valuation — a fourfold jump in under three weeks driven by compute-capacity pressure from an invite-only base that has passed 100,000 users, with founder Noah Shinn telling investors he wants the core product to stay free (Today's Startup News).

Illustration: an AI agent placing a phone call. This week both Instinct's Concierge and Meta's Muse gained the ability to call businesses on users' behalf. Illustration: TechPulse USA (AI-generated original)

Sometime this week, your AI assistant learned to use the phone. Not to set reminders or read you the news — to actually pick up the line, dial a number, and talk to a human on the other end on your behalf.

On Wednesday, Instinct founder Noah Shinn announced Instinct Concierge, a new feature that lets the startup's AI agent place calls: booking restaurants that don't take online reservations, getting you onto your dentist's cancellation list, sorting out a problem with your cable bill. Almost simultaneously, Meta gave its Muse agent the same power — the ability to place outbound calls to U.S. businesses.

Two rival agents, two phone features, one week. After years of AI assistants that could only type, the personal-agent wars just moved to the highest-trust channel in consumer technology: the telephone call. And the implications — for startups, for incumbents, and for anyone who has ever screened a call from an unknown number — are bigger than a feature launch.

What Instinct Concierge actually does

Instinct, the San Francisco startup reportedly valued at $10 billion, is one of the buzziest names in the personal-agent race. Its new calling feature, which Shinn announced on X on September 16, is pitched as a white-glove service for the tasks that still require an actual phone call.

"Your Instinct can now handle phone calls. Introducing Instinct Concierge – a white glove service meant to handle high-touch cases, such as making phone calls, high-end service booking, and more." — Noah Shinn on X, Sept. 16, 2026

The examples Shinn gave are telling, because they are the exact chores that have resisted every previous wave of automation. The restaurant that doesn't take online reservations. The dentist's cancellation list. The cable company. These are not glamorous tasks — they are the tedious, hold-music-laden errands of adult life, and they share one trait: they can only be resolved by a voice on a phone line.

The feature is rolling out in early access to some users, with plans to expand to all users over time, Shinn said. That's a meaningful qualifier — Instinct's user base has grown fast enough that the company has at times told users it was running at full capacity, and voice calls are far more expensive to serve than text. Early access is as much a capacity decision as a product one.

What Muse's calling does — and the odd way Meta launched it

Meta's Muse, which launched in the U.S. on September 8 as the centerpiece of Mark Zuckerberg's "personal superintelligence" push, also just gained the ability to place calls to U.S. businesses. But Meta's rollout mechanism is unusual: the feature is first available to users who ask Muse to make a call — essentially turning each request into a feature request.

Meta says that whenever Muse launches new features, they will be available first to people who ask Muse about them. It's a clever distribution trick for an agent that lives inside an app and WhatsApp: instead of announcing features and hoping users discover them, Meta lets demand surface itself. But it also means that, for now, Muse's calling exists in a gray zone — officially live, but only for the users who think to ask.

That caution fits the product's history. Muse, known internally as Hatch and modeled on the open-source agent OpenClaw, was delayed from April to September while Meta worked on safety; vice president of AI products Vishal Shah said the extra work let the company "cross the threshold" on its minimum requirements for safety, security, and privacy. Reuters reported that even after the delay, internal testers were encountering reliability problems, including one case where the agent allegedly exposed private iCloud photos. Giving that system the ability to call businesses on your behalf — to represent you, in your voice-adjacent name, to real humans — is a meaningful escalation of trust. Meta knows it.

Meta Muse Calling Rules: Verified Businesses Only (For Now)

Meta's calling test comes with tight guardrails — at least for now. In a private test reported by RuntimeWire, Muse limited calls to verified U.S. business numbers: it refused repeated requests to call a personal number and declined a one-time demonstration call because the tester's account did not have a confirmed number eligible for the option. Those restrictions matter because an agent that can browse the web and draft messages creates a different risk profile once it can place calls. Muse's terms prohibit bulk calls or messages without the consent required under telecommunications and marketing laws, including the Telephone Consumer Protection Act (TCPA). Limiting the test to businesses gives Meta a narrower surface on which to evaluate identity, consent, call quality, and failure handling — and keeps the feature on the right side of the robocall rules for now.

Why the phone is the real prize

Step back, and this week's news is about something larger than two feature launches. The personal-agent race — Instinct, Muse, Wajo, Town, Ollie, Poke, and a half-dozen others — has been a race to be the layer between you and your tasks. Text was the first battlefield: email, calendar, messaging. But an enormous share of real-world commerce still runs on voice. The holdout restaurants, the local tradespeople, the government offices, the customer-service lines — the long tail of the economy does not have APIs. It has phone numbers.

Whoever can reliably act through a phone call unlocks tasks that no app integration can reach. Voice has become the newest battleground — from Google's Gemini 3.8 Live voice AI to these calling agents. That's why several of Instinct's rivals had been touting calling as an advantage over Instinct — and why, as TechCrunch noted, Instinct's Concierge rollout brings everyone to parity. The phone was the last moat. Now it's a table stake.

The economics explain the timing. Instinct raised a Series B in late August that priced the company around $2.5 billion, and it is reportedly in talks for a new round at roughly $10 billion — a fourfold jump in under three weeks. That kind of money doesn't flow to a chatbot; it flows to a company that looks like it could become the default way tasks get done. A concierge that makes calls is a much easier story to tell than a text box.

The robocall problem nobody wants to talk about

There is a darker side to all of this, and it deserves more than a footnote. The telephone is a high-trust channel precisely because a voice on the line has historically meant a person. When AI agents start placing calls at scale — booking tables, haggling with cable companies, calling dentists — the line between "helpful assistant" and "robocall" gets very thin, very fast.

The questions pile up quickly. Does the restaurant know it's talking to an AI? Does the agent disclose itself? What happens when two agents call each other — your Instinct negotiating with the restaurant's own AI receptionist? And what stops this infrastructure from being repurposed for spam, scams, or the kind of social-engineering attacks that already cost Americans billions a year?

None of the announcements this week answered these questions in detail. Meta has said Muse operates with user-chosen app connections and a monitoring system. Transparency is the hard part — as OpenAI's own disclosure of its models' scheming behavior showed. And Instinct describes Concierge as a white-glove service — but neither company has laid out, in public, exactly how its agent identifies itself on a call or what safeguards prevent misuse. The U.S. already has robocall regulations built for an era of human telemarketers and recorded messages; AI agents that hold real-time conversations sit in a regulatory gray area that regulators have barely begun to map.

The industry is already building guardrails — including independent audits for AI agents, the "SOC 2 for AI agents" standard that just raised $40 million — but none of this week's announcements spelled out how calling agents fit into them.

This is the bet the whole industry is making: that consumer trust in AI voices can be built faster than the scams can erode it. It's a bet with real precedent on both sides — voice assistants like Siri and Alexa normalized talking to machines, while robocalls made Americans deeply suspicious of unknown numbers. Which instinct wins will decide whether agent calling becomes a utility or a nuisance.

Instinct Concierge and Muse Calling FAQ

Is Instinct Concierge available now? It is rolling out in early access to some users, with plans to expand to all users over time, founder Noah Shinn said. Instinct has at times told users it was running at full capacity — voice calls are far more expensive to serve than text, so early access is as much a capacity decision as a product one.

Does Muse tell people on the other end that it's an AI? Neither company has laid out in public exactly how its agent identifies itself on a call. Meta says Muse operates with user-chosen app connections and a monitoring system, but the disclosure mechanics remain undisclosed.

Can Meta's Muse call personal numbers? No — at least not in the current test. Muse told a private tester it could dial verified U.S. business numbers rather than personal numbers.

How much does Instinct cost? Shinn has told investors he wants to keep the core product free rather than introduce subscription pricing, pointing to advertising as a possible longer-term path to revenue. Concierge is positioned as a white-glove, higher-touch tier.

What to watch

The next milestones are practical, not visionary. First: how fast Instinct moves Concierge from early access to general availability, and whether call quality holds up at scale. Second: how Meta expands Muse's calling beyond the ask-first rollout — and whether it discloses the agent's identity on calls. Third: whether any regulator, or any major restaurant or hospitality group, pushes back on undisclosed AI callers.

The deeper trend is the one the Arkoselabs "Population Two" essay flagged this month: autonomous agents are arriving at consumer volume, not as demos but as daily infrastructure. The phone call was the last task that required a human voice. As of this week, it doesn't.

Analysis: the feature details above are from the companies' announcements and TechCrunch's reporting; the interpretation of what they imply is TechPulse USA's own. We have not independently tested either calling feature.

Sources

Feature announcements: TechCrunch (Sept. 17, 2026), Noah Shinn on X (Sept. 16, 2026). Muse launch and background: Reuters (Sept. 8, 2026), SiliconANGLE (Sept. 8, 2026). Instinct funding and valuation: TechCrunch reporting; Today's Startup News (Sept. 2026). Agent-race analysis: Arkoselabs "Population Two" essay (Sept. 2026). Related TechPulse USA coverage: OpenAI tests sponsored agents in ChatGPT.