DoorDash sees only "several hundred" AI agent orders

Silicon Valley has been hyping up agentic commerce as the next frontier where you never have to pick up your phone to buy anything again. DoorDash just gave that hype a reality check.

During DoorDash's Q3 earnings call on October 30, CEO Tony Xu disclosed that third-party AI agents like Muse, Instinct, and Fo have generated only "several hundred" orders in total. For scale, the delivery giant logs an average of 3 million deliveries every single day according to its 2025 report. In other words, agent orders don't even register as a rounding error.

Don't fight gravity, but don't be delusional

When asked about the space by an analyst, Xu struck a pragmatic tone. His take on technology is that you never want to fight gravity, but you also don't want to be delusional.

Delusional, in this case, means pretending people have stopped using apps. The overwhelming majority of users are still happily opening DoorDash on their phones to scroll menus and pick their meals.

Still, DoorDash isn't ignoring the space. Xu noted the company wants to understand emerging use cases, build healthy relationships with third-party bot makers, and perhaps even monetize them down the road by selling ads inside those agents.

DoorDash has also been aggressively shipping its own in-house AI tools. Back in June, it launched the Ask DoorDash chatbot, letting users order meals using natural prompts and food photos. In July, it introduced a CLI agent. In September, it debuted a text-to-order agent, alongside an MCP connector that allows corporate users to place catering orders directly from other chatbots.

Across the retail industry, attitudes toward bots remain wildly split. While Amazon actively blocks automated personal shopping bots from scraping its site, Walmart has laid out a red carpet for them, hoping agentic traffic might siphon business away from rivals.

Why it matters

This is a needed dose of reality for the entire agent ecosystem.

For everyday consumers, delegating dinner to a bot simply isn't worth the trouble yet. Ordering food is a visual, contextual, and often spontaneous decision filled with coupon stacking and dietary preferences. If an agent gets your burrito order wrong or picks the wrong delivery address, the friction of fixing it outweighs the five seconds saved. Sticking to the standard app remains far superior for now.

For the tech industry, DoorDash's stance offers a masterclass in hedging. Tony Xu acknowledges the potential direction of the technology without buying into the premature hype. By launching its own MCP and experimenting with conversational tools, DoorDash keeps control over its data and checkout pipeline while exploring potential ad revenue inside external bots.

We believe agentic commerce will eventually carve out real transaction volume, but it won't happen simply because bots exist. Until AI agents can genuinely handle real-time inventory, payment security, and personal taste better than a human scrolling an app, they will remain a tech demo rather than a shopping habit.