Restaurant websites are brochures: pretty, static, and mute. They show you the food, then hand you a phone number. Sushi On Fire was built to kill that handoff. Say a roll out loud and it lands in the cart. Confirm with your face and the kitchen has an order number.
This is part five of our series on the AI work we have shipped this year. Sushi On Fire is a demonstration restaurant we created to prove the pattern end to end; the checkout is a simulated Apple Pay sheet that declares itself a demonstration, and everything else on the screen is the actual product as it runs today.
The Work
The brief was to make a restaurant site take an order three ways, voice, chat, and tap, and to make the paying part disappear into about four taps. No apps to install, no account to create, no phone call.
Read the Sushi On Fire case study or open the live restaurant, tap the microphone, and order something.
What We Built
1. The agents
Flame speaks over a live voice model and the chatbot types over the same function tools: add to cart, check out, book a table. Every confirmation comes from a real tool result, never a claimed one.
2. The commerce
A cart that survives reload and a path from menu to paid in about four taps, ending in a simulated Apple Pay sheet with a Face ID confirm. No typing anywhere.
3. The kitchen
A ticket board for the chef, modeled on merchant tablets: what is new, what is on the line, what is ready. One tap advances a ticket and a chime announces a new order.
4. The owner's CMS
The owner edits the menu, prices, and imagery from a browser. The agents pick up the changes because they read the same data the site renders.
What Worked
The discipline that made it work is the same one behind our dealer agents: the AI is only allowed to confirm what a tool actually did.
ways to order: voice, chat, and tap
taps from menu to paid, roughly
shared set of function tools behind every agent
When Flame says a Rainbow Roll is in your cart and your total is twenty dollars and fifty-seven cents, that sentence was assembled from a tool result. The kitchen board updates because the order exists, not because the model said so.
What It Means for Hospitality
Restaurants pay commissions to delivery platforms largely because their own websites cannot transact. A site that takes orders by voice and chat changes the math on every direct order.
The deeper shift is conversational commerce with guardrails: guests do not want to browse a PDF menu on a phone. They want to say the thing they want and have it appear.
What this means for you: if your website's answer to a hungry customer is a phone number, you are paying for that gap, either in commissions or in walked-away orders.
Where the AI Does the Lifting
Two agents, one truth: the voice agent and the chat agent share the same function tools, so there is exactly one code path that touches the cart. The AI supplies the conversation; the tools supply the facts; the simulated payment sheet supplies the proof that the whole loop closes.
The brochure site
- - Shows the food, hides the ordering
- - Hands hungry guests a phone number
- - Menu changes need a web developer
- - The kitchen never hears from it
The working restaurant
- - Voice, chat, and tap all reach the cart
- - Menu to paid in about four taps
- - The owner edits everything in a browser
- - Tickets land on the chef's board
What It Does for the Business
For a restaurant owner, this is direct-channel revenue: orders that used to leak to aggregators, captured on a site the restaurant owns.
For Agentify AI, Sushi On Fire is the hospitality pattern we deploy from: agents on shared tools, owner-editable content, and a kitchen surface that closes the loop.
Final Takeaway: Menus Should Listen
The restaurant website has been a brochure for twenty years because taking orders was hard. It is not hard anymore. The whole loop, spoken order to kitchen ticket, runs in a browser today.
The standard to hold any conversational commerce to: every confirmation traced to a tool result. Anything less is a chatbot making promises.
Open the live restaurant and say a roll out loud. Watch the cart, not the chat, to see why it works.
