Resy Just Suspended a Diner for Using an AI Booking Agent: What Agentic Reservations Mean for NYC Restaurants

This is the first properly instructive AI story in hospitality, because nothing about it is hypothetical. A named platform, a named restaurant, a real consequence. As reported on September 10, JC Bahr-de Stefano, a director at Better Tomorrow Ventures, connected an AI tool called Instinct to his Resy account to chase a reservation at 4 Charles Prime Rib, the West Village steakhouse where tables disappear within seconds of release. The agent checked availability every ten minutes through the day and every 0.4 seconds around the moment reservations typically opened, amounting to roughly 200 API requests an hour. Resy temporarily closed his account for conduct violating its terms of service, and an American Express message warned that a repeat could permanently close both his Resy account and his credit card. The account was reinstated two days later.

What Resy's position actually is

Worth being precise, because it is more nuanced than a ban on AI. An American Express spokesperson said Resy does not allow unapproved bots or third party agents to independently access or interact with the platform, on the grounds that unapproved automated activity could pose risks and jeopardize a fair reservation experience for diners. At the same time, guests can find restaurants and make reservations through Resy's integrations with OpenAI's ChatGPT and Anthropic's Claude.

So the line is not human against machine. It is approved integration against unapproved automation.

Why should a restaurant care?

Three reasons, in order of how soon they matter.

The first is that your booking data may already be less accurate than you think. If agents are polling your availability, some share of the demand pressure you see is software rather than appetite. A restaurant reading that as genuine scarcity might hold prices or seating decisions against a signal that is partly artificial. The second is fairness, which is a brand problem rather than a technical one. If your hardest tables are consistently taken by whoever runs the best software, your most loyal guests lose access to you. That is a slow way to damage exactly the relationships worth most. The third is discovery, and it is the opportunity rather than the threat. Approved AI integrations mean an assistant may now recommend and book restaurants directly. That makes your structured information, hours, menu, location, description, into the material an assistant reasons over. It is the same argument as our Google Maps guide, extended to a new interface.

What should a restaurant actually do?

Nothing dramatic, and four things worth doing this month.

• Check your booking platform's stance on automated access and whether it flags suspicious activity to you. If your hard tables release at a fixed time, know whether that release is being farmed.
• Hold a share of your best tables outside the public release, for your own list. That is the simplest defense against your regulars losing access to software.
• Get your structured information right, since it is now what an assistant reads when someone asks where to eat. Hours, cuisine, location, price band, menu as text rather than a PDF.
• Enforce your no show policy consistently, because agent driven bookings raise the risk of tables held speculatively rather than intentionally.

That second point is the highest value one, and it depends on having a list. A restaurant that can offer its regulars first access does not need to win a technology fight. Our guest data and CRM guide covers what makes a guest list usable, and our piece on repeat guest revenue shows what protecting those relationships is worth annually.

Is this only a problem for hard to book restaurants?

The polling behavior is that nobody runs an agent to get into a room with open tables on Thursday. The discovery half applies to everyone. An assistant answering a question about where to eat near a neighborhood is drawing on the same structured data whether you are booked out for months or half empty. For most restaurants, that makes this a data hygiene story rather than a security story. The competitive risk is not an agent taking your tables; it is an assistant not having enough accurate information about you to suggest you at all.

What does this say about the next two years?

That the interface is moving and the fundamentals are not. Guests used to search, then browse, then book. Increasingly, they will ask, and something else will search, browse, and book on their behalf. What that layer reads is structured information plus reputation signals, which is more or less what search has always rewarded, delivered through a different front door. The practical implication is reassuring for operators who have done the basic work. Accurate listings, a real menu, answered reviews and a guest list you own remain the assets. Marqii's 2026 report found 89 percent of operators are familiar with how AI is influencing restaurant discovery and only 49 percent have made it part of their strategy, with the most common barrier among non adopters being not knowing where to start. Starting looks like listing accuracy, not a new platform, which is the same conclusion in our margin protection guide about where cheap gains actually live.

What does an assistant currently know about you?

Worth finding out, and it takes ten minutes. Ask a few assistants where to eat for your cuisine in your neighborhood, then ask directly about your restaurant. What comes back tells you what the layer above search has absorbed. The failure modes are informative. Being absent entirely usually means thin or inconsistent structured data. Being described with old hours, a former chef or a dish you removed means the sources being read are stale. Being confused with another business with a similar name means your citations disagree with each other somewhere. All three are fixable with listing work rather than with new technology, which is the reassuring part. Marqii found the most common barrier among operators not acting on AI discovery was not knowing where to start, and this is where it starts.

The uncomfortable question

If a guest used an agent to get a table at your restaurant, would you want to know, and would you cancel it? Most operators have not thought about it, and the honest answer for most is that a filled table is a filled table. That position is defensible right now and gets harder if speculative booking grows. What is not defensible is discovering in a year that your regulars stopped getting tables, and nobody noticed why.

Access is a brand decision

The Resy case is being read as an AI story. It is really a story about who gets access to a scarce table, which has always been a brand decision restaurants make deliberately or by default. Software just made the default less fair. Deciding on purpose and reserving something for the people who show up for you is the whole response required. My Chef Social helps NYC restaurants get their structured information and guest access working together, so discovery finds you and your regulars still get in. Get in touch if you want an audit of what an assistant currently knows about your restaurant.