Making a dinner reservation or checking whether an item is still in stock usually means a phone call, time on hold and, depending on the business, a second call to confirm. Restaurants, medical offices, repair shops and local businesses often have no online booking system or real-time inventory information. The call is the only option.
That kind of friction has been a quiet, persistent problem. An emerging category of artificial intelligence is now being aimed at it. Rather than answering questions, these AI systems are designed to act on them, contacting businesses, completing requests and reporting back once the task is done.
Google (GOOGL) is testing a new feature in Gemini that would allow the assistant to call businesses directly on a user’s behalf. The trial covers making reservations, checking product availability and rescheduling appointments. A user would hand off the request and receive confirmation once Gemini completes the call.
What Gemini would do that a chatbot cannot
A standard AI chatbot responds to prompts. It can summarize, recommend and explain, but the user still has to act on whatever it returns. Agentic AI is built differently. It interprets a goal, makes decisions within defined limits and carries out actions across digital services or, in this case, over the phone.
Google’s business-calling feature puts Gemini in that second category. A user who wants a table on a specific night would not need to find the restaurant’s number, call it or wait on hold. Gemini would handle the interaction and return with the result.
The feature is still in testing, and Google has not said it will be available to all users soon.
Google and Meta are competing for the same users
Meta (META) has been building toward this moment too. The company has expanded its Muse AI agent into shopping, travel and grocery services. Chief executive Mark Zuckerberg has said Meta plans to keep Muse free for a large number of interactions while eventually taking a small fee from transactions completed through the platform.
More Google:
- Alphabet’s biggest AI fear may be fading
- Bank of America says Alphabet stock investors are missing the bigger signal
- Google stock price faces major AI test ahead of earnings
Meta has added retail integrations with PayPal, Walmart, Best Buy, Gap, Sephora, Wayfair and Expedia, giving Muse a commercial infrastructure to operate from.
Google is coming at the same opportunity from a different starting point. The company already operates services that consumers use daily to find businesses, read reviews, check locations and book appointments. Gemini can connect to that existing data rather than building commercial relationships from scratch. Whether that early advantage holds will depend largely on what happens when Gemini makes its first calls.
Why trust is the harder problem to solve
The technology may be the easier part. Convincing users to hand over real-world tasks to an AI system is a different challenge.
A user delegating a restaurant booking to Gemini needs confidence that it reached the right business, communicated the request correctly and returned accurate information. They also need to know what Gemini did on their behalf, with a record they can check and clear steps to approve anything significant before it happens.
Businesses face a parallel set of concerns. AI-generated calls at scale could overwhelm phone lines, and staff would need ways to identify and respond to them appropriately. A misunderstood cancellation policy or an incorrectly confirmed appointment creates a real problem for both sides, and Google’s name is attached to each one.
The clearest model for what can go wrong is an agent that makes an error the user cannot easily catch until after the fact. Getting that right in testing is one thing. Getting it right across millions of different businesses, request types and local phone systems is another.
Bloomberg / Getty Images
How Google’s advertising model could create tension
Google built most of its revenue on search advertising. If Gemini becomes the primary interface between users and businesses, that model will need to evolve.
The company could charge businesses for agentic services, offer premium subscriptions or use Gemini to drive transactions through its existing commerce ecosystem. Each option carries a version of the same question: how does Google monetize the agent without giving users reason to doubt whether its recommendations reflect their interests or a paying advertiser’s?
Users who believe Gemini is suggesting a restaurant because someone paid for the placement rather than because it is the right fit will stop delegating tasks to it. Search advertising has not had to solve that precise problem before. Users running a search know it is influenced by ads. Handing a decision to an agent is different. The expectation is that the agent is working for the user, not alongside an ad buyer.
What investors should keep in mind
The opportunity is real. If agentic AI becomes widely adopted, consumers may rely less on traditional search and individual business websites. A user who asks Gemini to find a restaurant and complete the booking does not visit the restaurant’s site or click a search result. More of the relationship between users and businesses flows through the agent.
Several risks sit alongside that opportunity. The technology can make costly errors, and each one creates a customer-service or reputational problem with the user’s name on it. Regulatory rules around AI disclosure, consumer protection and privacy could limit what the feature is permitted to do. Users may also be slower to delegate real-world tasks than the technology is ready to handle.
Meta and other rivals are moving at the same time, and first-mover advantages in this space are not permanent. An agent that completes tasks accurately earns delegation. One that moves quickly and makes frequent mistakes pushes users back to doing things themselves. For Google, the challenge is not just building the feature. It is making it reliable enough that users choose to use it more than once.
Related: Mark Zuckerberg says Meta found a way to make more money
