DoorDash says AI agents have placed only several hundred orders

DoorDash says third-party AI agents have generated only “several hundred” orders on its platform, a small early signal for a market increasingly built around the idea that software agents will soon handle routine online purchases.

The disclosure came from CEO Tony Xu during DoorDash’s third-quarter earnings call, as the company discussed whether customers are beginning to order through external agents such as Muse, Instinct, and Fo. DoorDash said users still predominantly place orders through its own app, while signaling that it is willing to work with outside agent developers rather than block them outright.

The comments offer a reality check on agent-led commerce. DoorDash reported in its 2025 report that it completes an average of about 3 million deliveries per day. Against that operating scale, several hundred orders from third-party agents represent a very limited share, although the company did not provide a precise percentage or a breakdown by agent, geography, customer type, or time period.

Adoption remains far behind the agent narrative

Xu described agent-based ordering as a “dynamic area” and cautioned against both resisting technological change and overstating its present impact. His comments suggest DoorDash is treating the activity as an emerging channel to study rather than a material source of demand today.

That distinction matters for AI builders and commerce companies. Much of the discussion around AI agents assumes that consumers will soon delegate product discovery, comparison, payment, and fulfillment to autonomous software. DoorDash’s disclosure indicates that, at least on its marketplace, those behaviors have not yet displaced the conventional app experience at meaningful volume.

The company’s figure is also narrower than the broader category of AI-assisted shopping. It covers orders attributed to third-party agents, not necessarily every order influenced by a chatbot, recommendation system, or other AI feature. DoorDash did not say how it identifies an agent-originated order or whether its count includes partially automated journeys.

DoorDash is building its own agent interfaces

DoorDash is not waiting for external agents to define the category. The company launched a text-to-order agent in September, introduced a command-line interface agent in July, and released Ask DoorDash in June. The latter allows customers to place orders using prompts and photos, according to TechCrunch’s reporting.

DoorDash has also released an MCP, or Model Context Protocol, for ordering from other chatbots in corporate-ordering scenarios. The move gives developers a structured way to connect conversational systems with DoorDash workflows while allowing the company to retain more control over authentication, ordering, and fulfillment.

This two-track strategy—developing first-party interfaces while accommodating selected third-party integrations—could help DoorDash learn which agent use cases produce genuine demand. It may also allow the company to capture more commercial value if agents become an important discovery layer. Xu said there could eventually be an opportunity to advertise within those agents, though that remains a future possibility rather than an established business line.

What the evidence does and does not show

The central adoption figure comes from Xu’s remarks on DoorDash’s earnings call, as reported by TechCrunch. It is therefore a company-reported operational disclosure, not an independently audited measurement of the entire agent-commerce market.

The comparison point—about 3 million daily deliveries—comes from DoorDash’s 2025 report. It provides useful scale, but it does not establish the exact share of orders placed by agents because “several hundred” is imprecise and the two figures may cover different reporting periods. The company also did not identify how many orders came from Muse, Instinct, Fo, or other named systems.

A separate 24/7 Wall St. headline framed the figure against nearly 993 million DoorDash orders, but the underlying article text was unavailable in the supplied evidence. That total should therefore be treated as a reported comparison rather than a fully verifiable figure for this analysis.

The evidence supports a clear conclusion: third-party agent ordering exists on DoorDash, but it remains small relative to the company’s core delivery business. It does not, by itself, show whether the channel is growing rapidly, producing higher-value customers, reducing support costs, or creating measurable advertising revenue.

Implications for builders and enterprise buyers

For AI product teams, DoorDash’s experience highlights the difference between demonstrating an agent workflow and achieving durable user adoption. An agent may be technically capable of searching menus, selecting items, and submitting an order, but customers still need reasons to change a familiar habit. Trust, payment authorization, substitutions, delivery updates, refunds, and accountability all become more important when the customer is not operating the marketplace interface directly.

The limited volume also raises an attribution problem. Marketplaces need to know which agent initiated an order, what permissions it used, whether the customer approved the final basket, and how disputes should be handled. Those requirements favor explicit APIs and protocols over uncontrolled browser automation. DoorDash’s MCP release suggests the company is testing a more structured path for integrations, particularly in workplace ordering.

For enterprises, the immediate opportunity may be narrower than fully autonomous consumer shopping. Corporate meals, recurring orders, and internal purchasing workflows have clearer budgets and repeatable rules. DoorDash’s corporate-ordering integration could provide a more manageable environment for agent experimentation than open-ended consumer commerce.

The competitive question is how platforms choose to respond. TechCrunch noted that Amazon has blocked some personal agents, while Walmart has taken a more welcoming approach. DoorDash’s comments place it closer to the cooperative camp, but with an emphasis on learning, access controls, and potential monetization rather than unconditional openness.

What to watch next

The next meaningful signals will be whether DoorDash discloses growth in agent-originated orders, identifies which integrations are driving activity, or reports a material difference in basket size, repeat use, or fulfillment performance. Clearer definitions of an “agent order” would also make future comparisons more useful.

Builders should watch whether DoorDash expands its MCP capabilities beyond corporate ordering, adds more transaction controls, or introduces advertising products inside third-party agents. The reception of Ask DoorDash, the text-to-order agent, and the CLI agent will indicate whether customers prefer DoorDash-controlled interfaces or external assistants.

More broadly, the market will need evidence that agents can generate incremental demand rather than simply reroute orders that customers would have placed in an app. Until that distinction is visible in platform data, claims about agents taking over commerce remain ahead of demonstrated usage.

Creati.ai perspective

DoorDash’s disclosure is important less because several hundred orders constitute a major business than because it supplies a rare counterweight to sweeping predictions about autonomous commerce. A leading delivery marketplace has exposure to real transactions, and its early third-party volume remains modest beside its existing app traffic.

The lesson for AI companies is practical: agent capability is only one part of a transaction product. Adoption will depend on permissioning, reliability, merchant and customer support, and a business model that benefits the platform as well as the agent maker. DoorDash appears to be positioning itself to learn without surrendering control, and the next phase will show whether that balance can turn experimentation into repeat usage.