On 30 October 2026, TechCrunch reported that Zest, a restaurant discovery app launched earlier this year, now runs an AI concierge over text messages. You text it, or add it to a group chat, and it suggests restaurants by cuisine, craving, occasion, dietary restrictions and neighbourhood. It learns your taste from your card transactions, linked through Plaid. In a group chat it blends everyone’s tastes, settles the argument with an in-thread vote, then checks availability and books through OpenTable, Resy or SevenRooms without anyone leaving the conversation. Zest was founded by Mario Gomez-Hall in November 2024 and has raised US$1.8 million in pre-seed funding.
The excitement is the group vote. The change is somewhere else.
The feature people will share is the voting. Six friends, one thread, no more forty messages about where to eat.
But I always ask what actually changed. Voting, recommendations and reservation systems all existed before. The new part is that nobody in that thread opened a browser, read a menu or called the restaurant. The agent took in transaction history, group preferences and reservation inventory, and sent a reservation out. The humans only said yes or no.
Every input and every output here is structured data, exchanged as text. No voice, no browsing, no reading a page the way a person reads it.
This is what agentic commerce looks like on a normal evening
My view on voice AI is that it helps on input and much less on output. I dictate to Claude with Wispr Flow, and for giving instructions it works. But a conversation that ends in a booking, a payment or a change to an order is increasingly an agent talking to another system. Agents will not talk to each other in voice. They exchange data, and data is text.
Zest is a small, plain example of exactly that. No stage demo. A text thread, a vote, a table.
If your availability is on a phone line, you were never in the decision
Think about the restaurant that never made the shortlist.
The agent picks from what it can read and book. If your tables sit on OpenTable, Resy or SevenRooms, you were a candidate. If reservations happen by phone call, Instagram DM or WhatsApp to the owner’s phone, the agent cannot check you or book you, so it does not show you. If your menu is a photo of a PDF on a Facebook page, the agent cannot read dietary options off it, so the vegetarian in the group rules you out without knowing you had three dishes for them.
For the big platforms, agentic commerce is still 50/50. Some will open up to agents and some will block them to keep their grip on shoppers. A twelve-table restaurant cannot wait it out. Wherever your data shows up, it needs to be precise, detailed and accurate about what you are. The attributes an agent matches against six people’s preferences are things like halal, vegan options, quiet, private room for 12, under S$40 a head, open on Monday.
And the questions come in a long tail. Somewhere quiet near Tanjong Pagar with good fish. Not too pricey. Can take a pram. To catch that intent, your information has to cover a broad spread of questions with specific attributes, not a tagline and a phone number.
The same test applies to clinics, salons and anyone who takes bookings
Swap the restaurant for a dental clinic whose diary lives on a phone line, a salon that books by DM, or a tuition centre that confirms slots by email two days later. The moment a customer’s agent wants to check a slot and commit, those businesses are invisible in the same way.
This is also the layer I build on at AeroChat. A customer asking on WhatsApp to book, pay a deposit or change a delivery is text and data that turns into an action in the business’s systems. The Zest story is the same thing seen from the customer’s side: an agent acting for them, reaching into whatever data the business has made reachable.
What to do now
- Find out where your availability actually lives. If the only way to check it is a phone call or a DM, no agent can book you. Put it in a system that exposes it.
- Put your menu or service list on your own site as text, with the attributes a person would filter on: dietary, price range, seating, hours, location, group size. Engines and agents quote specific facts and drop adjectives.
- Ask ChatGPT, Gemini and Perplexity the questions your customers ask, and note the date. If you are not named in the answers, fix the data before you spend on anything else.
- Do not wait for Zest to reach your city. Whichever agent your customers end up using, it will choose the same way.
Source: TechCrunch: Zest’s new AI agent finds restaurants and books reservations over text