- Third-party AI agents have generated only several hundred orders on DoorDash since the capability launched, against nearly one billion orders the platform handled in Q3 2026 alone.
- DoorDash’s “Ask DoorDash” conversational assistant yields baskets with nearly 50% higher value than non-AI orders.
- An internal DoorDash grocery reorder agent cut flow latency from 16 seconds to 2 seconds, while DashBuddy handles the bulk of new driver sign-up questions.
Several hundred. That is how many orders third-party AI agents have placed on DoorDash since the capability launched, set against nearly one billion orders the platform processed in Q3 2026 alone. It is the most honest data point in circulation about where agentic commerce actually stands.
Third-Party Agents Barely Register
DoorDash CEO Tony Xu called third-party AI agent orders a “fairly low use case” on the company’s October 29, 2026 earnings call. The “several hundred” figure covers orders routed through external services including Muse, Instinct and Fo, each of which aims to automate purchasing on behalf of users. DoorDash did not specify a timeframe for that cumulative count, but even over an extended period the number is negligible. Consumers still overwhelmingly open the DoorDash app directly when they want food, Xu noted.
What Blocks Agent Orders
Connecting third-party agents to real-time merchant inventories is harder than it looks. Menus change constantly, and agents working from stale data surface wrong prices and mismatched photos. That credibility problem gets worse when the agent also holds a stored payment method and can act without a confirmation step. Walmart and US banks are hitting similar friction as they push agents into payment workflows. AI power users tend to reach for agents on coding and automation tasks first; dining and reservations are well down the list.
Where DoorDash’s Own Agents Work
The internal grocery reorder agent is the clearest success. By precomputing recurring purchase needs, it cut flow latency from 16 seconds to 2 seconds, doubled shopping list views and lifted the flow’s order rate by about 34%. DashBuddy, an AI assistant for delivery drivers, handles the bulk of questions from new sign-ups, according to the company. Both are narrow, well-defined workflows built on fresh internal data, which is precisely why they perform where third-party agents do not.
Built for an Agent-First Future
DoorDash co-founder Andy Fang laid out the company’s longer-term position at the company’s Dash Forward 2026 event in San Francisco on September 29, 2026: DoorDash wants to be the execution layer AI agents reach for when something needs to move in the physical world. That means surfacing real-time retailer inventory, current discounts and courier availability to whatever agent is doing the ordering.
The company has already shipped a Model Context Protocol connector for external AI tools, though it remains limited to early beta partners and is used mainly for corporate lunch orders. Ongoing discussions with Meta over official Muse integration suggest DoorDash is serious about staying relevant as agent-driven commerce develops. Advertising inside third-party agents is floated as a future revenue line, but the current priority is understanding how developers actually use the infrastructure before monetising it.
Xu closed the Q3 call with a line worth keeping: he said his read on technology is to “never want to fight gravity,” paired with a warning against delusion about how fast things move. With several hundred agent-placed orders against nearly a billion human ones, both parts of that advice apply right now.
