A Y Combinator startup called VOYGR has tested how San Francisco businesses react when an artificial intelligence agent phones them directly. During SF Tech Week, the company used its own PlaceCall API to place more than 1,000 outbound calls to local businesses. The AI identified itself as an AI on every call, asked one question, and then waited for a reply. Only about one in three businesses kept talking.
The findings, shared in a company post that circulated on LinkedIn and was later reposted on Reddit, offer an early look at how ready small businesses are for AI-driven phone contact. The source material does not name an independent publication, so the figures below come directly from VOYGR's own account of the experiment.
How the experiment worked
VOYGR, a member of the Winter 2026 batch of Y Combinator, built the test around PlaceCall, its API for developers who want AI agents to make real telephone calls. The calls covered a wide mix of neighborhood services, including veterinary clinics, dentists, bars, cafés and hair salons.
The design was deliberately simple. Each call opened with a clear statement that the caller was an AI, followed by a single question meant to be useful to the business. VOYGR treated this upfront disclosure as a key variable, because it shapes how people respond in the first seconds of a call.
The company has also published a map showing every business that was called, along with a record of what each one said. That transparency lets readers check individual outcomes rather than rely only on summary numbers.
Big differences between business types
Response rates varied sharply by sector. According to VOYGR, veterinary clinics kept the conversation going 79 percent of the time, while dentists did so 61 percent of the time. Bars and dry cleaners were far less willing to engage, with only 20 percent staying on the line. In many of those cases, the person who answered said nothing at all and hung up roughly five seconds after the introduction.
The gap suggests that the nature of a business matters as much as the technology calling it. Clinics and dental practices handle appointments and enquiries all day, so a call that offers a clear, relevant question may feel like a normal part of their workflow. Bars and dry cleaners, which often have staff juggling customers in person, may see an unsolicited call as an interruption with little benefit.
AI calling AI
One unexpected result was how often the AI reached another automated system. VOYGR reported that roughly one in 25 of the businesses it called already uses an AI system to answer the phone. That means some calls became conversations between two machines, which raises practical questions about efficiency and about how these exchanges should be handled.
In at least two of these AI-to-AI calls, the answering system asked VOYGR's agent to say "zebra" to prove it was a person. The request is a playful nod to the old idea of a simple test for telling humans and machines apart. It also highlights how quickly automated phone systems are learning to screen incoming calls.
VOYGR also said that 94 percent of the San Francisco dentists who answered are taking new patients. The company presented that figure as a sign of demand in a sector where appointment availability often matters to patients, although the post did not describe how the number was calculated in detail.
Why businesses are not ready yet
The company argues that the experiment points to a coming shift. Personal assistant tools such as Instinct, Muse and Dots are expected to start calling businesses on behalf of users to book appointments, ask questions and check details. According to VOYGR, the current reaction is far from welcoming. Right now, two out of three businesses simply hang up.
That reaction reflects a wider tension in consumer technology. Many owners of small businesses are still deciding whether automated calls are a helpful service or an unwanted intrusion. Their answers will likely depend on how transparent the calls are, how relevant the questions feel, and whether the caller respects the time of the person who answers.
Industry observers have also begun to discuss whether AI callers should always disclose their identity. VOYGR's approach, which made disclosure a standard part of every call, offers one model. Whether that becomes a common expectation will depend on user experience, customer trust and the rules that governments may choose to adopt.
What comes next
The VOYGR experiment is a small, self-reported study, and it does not show how these calls would perform over months or across different markets. Still, it gives a concrete picture of the friction that AI phone agents will face as they move from demonstrations into everyday use. For businesses, the lesson is that automated calls will become more common, and how they are designed will shape whether they are welcomed or ignored. For developers, the results suggest that the first few seconds of a call may matter as much as the technology behind it.
Related: LFM2.5-24B-A2B: Liquid AI's Open Reasoning Model for On-Device Agents · Awesome Jev Catalogs 859 Projects for Fast Agent Decision Routing
Comments (0)