DoorDash processed 970 million orders in its second quarter this year. The haul generated $4.5 billion in revenue. That works out to roughly $4.64 per order. The number matters. It funds drivers, insurance, support and the marketplace that connects hungry customers to restaurants.

Yet in August the company sent a blunt email to Bay Area restaurants. Some might have been listed on Bites without permission. Bites. A 10-person pre-seed startup with only about 300 partner restaurants. The warning carried weight. Even a blip drew DoorDash’s full attention. The Verge reported the exchange.

From ChatGPT Orders to Text Messages

Bites sells itself as AI native. Diners type requests straight into ChatGPT. The order travels directly to the restaurant through point-of-sale integrations. No marketplace intermediary. No complex fee structure. Just a flat $1 delivery charge. Diners usually pay it. Some restaurants cover the cost themselves.

Simpler. Cheaper. At least on paper. A Verge test ordered pizza and a drink. Total hit $56.73 with tip, tax and fees. The same order through traditional platforms ran about $15 higher. Restaurants click one button to opt in once their POS provider partners with Bites.

DoorDash responded carefully. Spokesperson Utkarsh Singh told The Verge the information made clear no action was needed for opted-in partners. The email simply flagged potential unauthorized listings. Yet the episode reveals anxiety. AI agents promise to handle tasks like ordering food. They could erode the discovery, advertising and data advantages that platforms guard.

DoorDash refuses to sit idle. On Sept. 30 it unveiled its own text-to-order AI agent inside Apple’s Messages app. Co-founder Andy Fang described it at the Dash Forward event in San Francisco. Users text prompts such as “order my usual protein bowl to the office.” The agent matches the phone number to a DoorDash profile. It pulls order history and preferences. Then it builds a cart, shows photos pulled from restaurant pages, and completes checkout with stored payment details. Bloomberg detailed the mechanics.

The feature sits in limited beta for about 20,000 U.S. users. Others join a waitlist. Early testers report it grows smarter with repeated use. It learns habits. It suggests new spots. It even handles group orders with mixed dietary needs and quantities in a single thread. Send a photo of an empty fridge. The agent identifies missing ingredients for a recipe and arranges delivery. Fang noted his personal agent texts him weekday mornings to ask about lunch. Otherwise he forgets and ends up hungry by mid-afternoon. TechCrunch covered the launch.

But imperfections remain. Bloomberg’s initial tests found price discrepancies between the agent’s quotes and the actual app. Photos sometimes failed to match recommendations. DoorDash says waitlist users receive an improved version. These hiccups highlight the gap between demo and daily reliability. Still, the direction feels clear. DoorDash built the Messages agent on technology from its June in-app chatbot called Ask DoorDash.

And the company pushes further. It introduced a connector for corporate AI tools. Teams at SpaceXAI, Vercel, Cognition, Tempo and Mercor already test it for office lunches and supply restocking. Employees stay inside Slack or internal agents. The system searches, carts, orders and tracks. No app switching. DoorDash also plans drone delivery tests in Northern California with partners like Chipotle. Early flights average under five minutes.

CEO Tony Xu addressed the broader agent threat directly in recent comments. AI shopping assistants might manage initial discovery and ordering. Yet they still need someone to fulfill the physical delivery on time and in good condition. DoorDash owns the end-to-end piece. That integration, he argues, creates the advantage. Restaurant Business Online quoted Xu.

The tension runs deeper than one startup versus one giant. OpenAI, TikTok and Pinterest experiment with agents that guide users straight to purchase. Each seeks a slice of that final transaction. For food delivery the transaction carries high marginal costs. Drivers, packaging, timing. Agents compress only the digital front end. The physical logistics stay expensive. This reality suggests fee pressure more than outright replacement.

Restaurants with their own delivery fleets stand to gain. They capture demand without marketplace cuts. Platforms that rely solely on the app surface face margin squeeze. Bites remains tiny. Its model requires scale to matter. Yet its existence forces bigger players to accelerate. DoorDash’s rapid response, from text agents to enterprise connectors to drones, shows how seriously it takes the shift.

Fang posted on X that agents represent a new way to interact with apps. The statement carries confidence. DoorDash already dominates U.S. food delivery. Its moves aim to keep that position as AI rewrites the rules of ordering. Diners may soon skip apps entirely. They will simply text what they want. The platform that answers fastest, most accurately and most profitably will win the orders. For now DoorDash intends to be that platform. But the small challengers keep knocking.

Industry watchers note the pattern. Last quarter’s numbers prove the prize. Any agent that books dinner competes for part of that $4.64. The real battle may not eliminate DoorDash. It could instead compress the software layer while logistics costs hold steady. The outcome will shape not just food delivery but the wider agentic commerce push. Restaurants, platforms and AI developers all watch closely.