The phone call nobody wanted to make — so an AI voice agent makes it
There is a special category of dread reserved for calling a restaurant to ask how long the wait is. The hold music. The harried host shouting over the kitchen. The 45 minutes you suspect nobody actually measured. Yelp’s new “Check for Me” feature proposes a refreshingly direct answer: stop calling. Tap a button on the restaurant’s page, enter your party size and phone number, and an AI voice agent makes the call for you. The wait time arrives by text.
It is a genuinely clever piece of consumer convenience, and it solves a real annoyance. But consider what happens at the other end of that call. The phone number the agent dials belongs to a restaurant that may itself have stopped answering its own phone.
Meanwhile, on the other end of the line
Yelp has been building the business side of this conversation for a while. Last year, it debuted a voice solution for restaurants and service providers — software designed to sit on the business end of the line.
Then there is the newest paid addition to the stack. Smart Lead Reply, at $30 per month, is an AI agent that replies to customers and engages with them to qualify a lead; business owners can specify how the agent should handle a conversation. The description covers replying to customers and qualifying leads — not handling phone calls, and nothing in the launch coverage says what Check for Me itself costs.
The part the launch coverage couldn’t help but say
Here is the delicious detail, and it comes from TechCrunch’s coverage of the launch — not from Yelp itself: “there is a possibility that a user’s agent will talk to the restaurant’s agent, and agent-to-agent communication might break down the conversation.”
Read that again. Yelp is shipping a consumer agent that places calls to restaurants, while it already sells a voice solution that handles calls for restaurants and service providers. The two-sided scenario is not a conspiracy theory; it is the stated, open possibility in the launch coverage. But note what that means — and what it doesn’t. The coverage names agent-to-agent calls as a possibility; it does not establish whether those calls are already happening or how common they are. That distinction — a reported possibility versus a demonstrated deployment — matters, because the questions the trajectory raises are unanswered either way: if a Yelp consumer agent calls a restaurant whose phone is answered by another agent — two pieces of software with conflicting instructions, say, around waitlist priority — what happens? Who arbitrates? And what recourse does the diner have if the two programs fail to understand each other?
The collision is structural. The agent-to-agent phone call is the open possibility the coverage names; Yelp just happens to be building the road from both directions at once.
The disclosure question nobody can answer
Which leaves one question rather pointed. For Yelp’s restaurant voice solution — the product known to handle calls — does it disclose to the caller that they are speaking with an AI agent? The evidence is silent, and the question matters precisely because the caller on the other end may soon be an AI too.
To be fair to Yelp
The steelman case is strong, and it deserves its paragraph. Restaurant owners genuinely miss calls and lose bookings during rush hours; diners genuinely hate calling ahead. Both products are positioned as voluntary — a restaurant that wants only human contact can, in principle, keep answering its own phone. “Yelp talking to itself” may overstate today’s reality: the agent on the diner’s behalf may reach a human host. And the breakdown scenario — two agents failing to complete a call — may be a failure mode the industry is actively engineering against, not evidence of some self-dealing scheme. In principle, nobody is forced into the loop.
But notice what the steelman concedes: the loop is being assembled. One company now offers the agent that rings the phone and the software that answers it, and offers the consumer side as pure convenience. The economics of each product make sense in isolation. The interesting part is what happens when the isolations collapse — when the consumer convenience and the business automation finally meet on the same call.
The questions nobody has answered yet
How would agents from different vendors complete a call if both ends were software with conflicting instructions? And what does Yelp charge — or plan to charge — on each side when both ends of a call run on its own agent products? The coverage says nothing about the cost of Check for Me, and nothing about two-sided monetization of the call itself.
Until those answers arrive, the amused unease is the appropriate response. But the coverage itself raised the scenario, so let’s ask it plainly: one Friday night, what if the number your agent dials is answered by another agent — two programs with conflicting instructions around waitlist priority, trying to complete a call that neither diner nor host is part of? Who arbitrates if they fail to understand each other, and what recourse does the diner have?
Should that call ever happen, the first casualty of the agent era may not be a job at all. It might be a dinner plan. The future of calling a restaurant, it turns out, may not require calling at all. It may only require hoping your agent and their agent can hold a conversation.
