Skip to main content
Back to News
news/AI Applications

ChatGPT Ordering Threatens Food Delivery's Middleman Model

Diners can now order food inside ChatGPT, skipping DoorDash commissions. The fight over AI agents that buy on your behalf is reshaping delivery economics.

Stefan Trbojevic

Stefan Trbojevic

7 October 20262 min read
LinkedIn
Abstract orange network of routing nodes connecting a central AI hub to storefront-like blocks

The takeaway

Agentic commerce turns delivery platforms from gatekeepers into optional middlemen; the battleground shifts to POS integrations, agent identity and customer data ownership.

Why it matters for builders

If agents become the buying layer, integration work moves to POS/menu data, agent authentication standards and data custody. Builders should design for merchant consent and identity, not for app-level discovery.

ChatGPT Ordering Threatens Food Delivery's Middleman Model

A ten-person Bay Area startup is trying to do what delivery platforms have feared for two years: let an AI agent order dinner without the platform in the loop. Bites, a pre-seed company co-founded by Grubhub co-founder Matt Maloney, lets diners order from roughly 300 restaurants directly inside ChatGPT, with orders routed straight to the kitchen through point-of-sale providers like Toast and Square. In August, DoorDash responded by mailing Bay Area restaurants a warning letter.

What happened

DoorDash's form letter, reviewed by The Verge, told restaurants they may be listed on Bites without their consent, that they may not have agreed to Bites' "terms, pricing structures, or service standards," and that "depending on the state, such practices could be illegal." DoorDash spokesperson Lisa Singh said "some restaurant partners raised concerns" about appearing on the platform without opting in.

The economics are the real fight. DoorDash charges restaurants 15 to 30 percent per delivery order; Bites charges a flat one-dollar surcharge and pulls digitized menus from point-of-sale systems straight into an LLM. The Verge tested the same pizza and beverage order twice: $56.73 through ChatGPT and Bites, $70.29 on DoorDash, with the identical pizza priced $11 higher.

INLINE_IMAGE_HERE

Abstract diagram of AI agent order routing bypassing a central platform node

Why it matters

If large language models become the portal to commerce, platforms lose the thing they actually monetize: the customer relationship. DoorDash's advertising business reached $1 billion in 2024, and Amazon's ad business generated $19.8 billion in a single quarter, driven by sponsored product listings. An agent that already knows which restaurant a user wants never sees a boosted result.

That tension is already in court and in product roadmaps. Amazon sued Perplexity in 2025 over agentic browsing, has blocked Meta's Muse agent from shopping on its site, and joined Meta, Walmart and Stripe in publishing a "personal agent protocol" for how bots authenticate with businesses. DoorDash itself said in September that it is building iMessage ordering by prompt.

What it means for builders

Agentic commerce is becoming an integration problem, not a model problem. The rails are point-of-sale connections, menu data schemas and agent identity standards, and whoever owns those rails owns the order. Builders wiring agents into commerce should watch three things: how the emerging personal-agent protocol handles authentication and merchant consent, who keeps custody of customer data once an agent rather than an app holds the relationship, and whether discovery survives when users arrive already knowing what they want.

Discovery may be the last moat. Bites founder Bala Subramaniam concedes the startup "may not become the discovery platform." Delivery apps spent years and billions building exactly that. Whether it is still worth anything in an agent-first internet is now an open question.

Share𝕏

The Automation Brief

Read 5 AI stories instead of 50.

The essential moves in AI agents, models, automation and infrastructure — filtered for builders and operators, with the part that actually matters.

No noise. Unsubscribe anytime.

Editorial notes

Reported by

Stefan Trbojevic

Edited by

n8n Lab Editorial

Published

7 October 2026

Updated

7 October 2026

AI disclosure: AI assisted with research and drafting. Factual claims are reviewed by an editor.

n8n Lab is an independent service provider. We are not affiliated with, endorsed by, or sponsored by n8n GmbH. “n8n” is a trademark of n8n GmbH and is used here only to describe the platform-specific implementation and automation services we provide.