The next disruption in restaurant ordering may not arrive as an app. It may arrive as an agent. That is the premise of a new analysis from Restaurant Business Magazine, which asks how AI agents — software systems that act on a customer's behalf, browsing, comparing and completing purchases without human clicks — could change the way food orders reach the kitchen.

The piece lands at a moment when the industry has already absorbed one round of ordering automation. QR codes, kiosks and app-based checkout moved the transaction off the cash register. AI agents propose something different. The customer delegates the decision itself. The software chooses the restaurant, assembles the order and pays.

What would an agent actually replace?

The distinction matters for operators. A kiosk replaces a cashier at the point of sale. An agent replaces the discovery and selection layer that currently belongs to Google, DoorDash, Uber Eats and the chains' own first-party channels.

Who controls that layer controls the commission. If agents transact on a marketplace's rails, restaurants keep paying the aggregator fees they have spent years trying to escape. If agents transact directly with operator systems, the economics shift back toward the house.

Restaurant Business frames the question rather than settling it. That restraint is fair. The technology is young, and the commercial plumbing — machine-readable menus, real-time availability, programmatic payment — is uneven across the sector.

Who pays, and for what?

For technology stories the arithmetic is the story. Agent-mediated ordering implies costs on several fronts:

• Integration spend, so an AI system can read the menu and confirm the order
• Menu data discipline, since an agent cannot upsell what it cannot parse
• Possibly new placement fees, if ordering agents become a bidding surface the way search and marketplace rankings already are

Independent operators with thin margins and little technical staff would feel this first. Large chains, which already maintain structured menu data for apps and kiosks, are better positioned to meet an agent halfway.

The menu engineering angle

There is a quieter consequence worth flagging. When a human scans a menu, design does persuasion — photography, placement, price anchoring. When an agent scans a menu, persuasion becomes a data problem. Item names, modifiers, portion logic and price must be legible to a machine that has no appetite and no impulse control.

Menu engineering (the practice of steering mix toward high-margin items) would migrate from layout to metadata. An item invisible to the agent is an item that does not sell. Expect vendors to sell structured-menu tooling on exactly that fear.

A wait-and-see verdict

Restaurant Business does not predict a timeline, and it is right not to. Delivery marketplaces took a decade to reach their current penetration, and they solved a simpler problem: they moved an order that already existed. Agents attempt to move the decision before it exists.

For now the rational operator move is unglamorous. Keep menu data clean. Watch the marketplace announcements, because the aggregators have the most to lose and the most engineers to spend. The counter will look the same for a while yet. The order behind it may not.