DoorDash Says AI Agents Have Generated Only Hundreds of Orders

Introduction

AI agents are often presented as the next interface for commerce. Instead of opening a delivery app, a customer could simply describe a meal, a budget, or a set of preferences and let an agent search, choose, and place the order. DoorDash’s latest disclosure, however, suggests that this transition is still far from established.

Key points

• CEO Tony Xu said during DoorDash’s third-quarter earnings call that the company has seen only “several hundred” orders in total from third-party agents, including Muse, Instinct, Fo, and others.
• DoorDash reported in its 2025 materials that it completes an average of about 3 million deliveries per day. Against that baseline, third-party agent orders remain a very small volume.
• Customers still predominantly order through DoorDash’s own app. The company nevertheless wants to learn more about agentic use cases and build a healthy relationship with external agent developers.
• DoorDash is also developing its own agent experiences, including a text-to-order agent, a command-line agent, and the Ask DoorDash chatbot. It has released an MCP that enables ordering from other chatbots for corporate use.

Why it matters

The disclosure highlights a gap between the attention surrounding AI agents and their current consumer traction. An agent may be able to interpret a natural-language request, but that does not automatically mean customers will trust it with payment, addresses, substitutions, delivery instructions, and support. Reliability, authorization, platform access, and accountability all have to work together before agent-led ordering can become routine.

DoorDash’s position is also more nuanced than simply accepting or blocking agents. Xu argued that companies should not fight technological momentum, but should also avoid losing touch with present-day reality. DoorDash is watching external agents for signs of genuine demand while building its own tools to retain control over the ordering experience.

The wider web is still deciding how agents should access commercial content. The source notes that Amazon has been blocking personal agents, while Walmart has taken a more welcoming approach in an effort to capture business. These choices reflect a strategic trade-off. Opening the door to agents may create an additional sales channel, but it can also weaken a company’s direct relationship with customers, reduce control over presentation, and change the economics of advertising and data.

DoorDash has suggested that advertising could eventually become part of agent interactions. That would make agents not only transaction interfaces but also new distribution layers for commercial messages. Yet with third-party activity still measured in only hundreds of orders, the immediate priority is validation: whether users return, whether transactions are dependable, and whether the parties can establish clear rules for consent and responsibility. The current evidence supports cautious experimentation, not the conclusion that agents have already taken over commerce.

Source: TechCrunch AI