Meta privately tests Muse agents that can call verified US businesses

Why it matters

Phone calls move Muse into customer service and offline commerce, while the demonstrated business-only limits highlight that telephony brings immediate consent, identity and abuse risks.

Meta is privately testing a Muse capability that lets agents call verified US businesses, according to posts from Meta employee Matt Schlicht and founder Matt Van Horn. The exchange shows how Van Horn gained access through a chain of Meta employees, not that Meta has released phone calling broadly.

Schlicht said he used his Muse agent to call Tesla dealerships because his wife wanted a Tesla Model X and he wanted to test one before proceeding. The agent contacted dealerships to arrange an appointment, turning a task that would ordinarily require searching for locations and calling sales staff into an agent workflow.

Van Horn replied that his own Muse agent could not place the same calls. Schlicht then tagged another Meta employee, who brought in another colleague, and Van Horn subsequently received access to the capability while it remained in private testing.

Van Horn (@mvanhorn), a founder and prolific agent-tool builder, posted what he described as an early-access alert on X on September 16th. He credited Schlicht (@MattPRD), Ryan Fox (@wailord) and Darren Scerri (@darrenscerri) for getting the calling test onto his account.

Van Horn previously co-founded Zimride, the carpooling service that became Lyft, and later co-founded connected-appliance maker June. Weber acquired June in January 2021. His recent work has centered on practical agent infrastructure and the ways non-engineers can assemble automated workers with existing models and tools.

Van Horn said he had already added phone calling to his Grok Bot setup using Twilio. In the Muse test, calling appeared to be handled directly by Muse, without requiring users to connect and manage a separate telephony provider.

The evidence remains limited to a private test on selected accounts. Van Horn said he was still trying to hear the experience from the call recipient's side and answered "Hopefully soon" when asked when the capability would become available.

A screenshot of Van Horn's exchange with Muse shows the current boundary. Muse refused repeated requests to call Van Horn's own number, saying the test could dial verified US business numbers rather than personal numbers. Muse also told him that a one-time demonstration call was unavailable because his account did not have a confirmed number eligible for that option.

Those restrictions matter because an agent that can operate a browser and draft a message creates a different set of risks 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. Limiting the test to businesses gives Meta a narrower surface on which to evaluate identity, consent, call quality and failure handling.

Schlicht joined Meta Superintelligence Labs after Meta acquired Moltbook, the experimental social network for AI agents that he created with Ben Parr. Axios reported in March that Schlicht had been working on autonomous agents since 2023 and was scheduled to start at Meta on March 16th.

The employee chain behind Van Horn's access connects the calling work to Meta's broader effort to put acquired founder talent into Muse's action layer. Moltbook explored how autonomous agents identify themselves and coordinate with other agents. Phone calling pushes a related problem into the existing telephone network, where an AI worker must interact with people and businesses that never agreed to join an agent platform.

Meta launched Muse on September 8th as a personal agent that can operate its own browser, connect to apps, fill out forms, send emails and make purchases with user approval. Muse runs on a dedicated virtual machine, while a separate Sentinel agent is supposed to review actions before they reach the internet and request permission for sensitive steps.

Calling is consistent with that product direction. Muse is designed to continue working after the app closes, handle multiple tasks and build tools when a job requires them. Phone access would extend the agent into the large portion of commerce and customer service that still ends with a person answering a number.

It would also require Muse to select the right number, remain within the user's instructions, manage disclosure of its machine identity and avoid turning inexpensive automated labor into a robocalling system. Each call could produce sensitive conversational data that would need to fit within the controls Meta has promised for Muse's virtual machines and activity logs.

Van Horn's Twilio workaround showed that determined users could already attach calling to an agent. The private Muse test suggests Meta is working to make that capability native, while its account-level access and business-only restrictions show that the company is not yet treating phone calls as a routine agent action.