Few AI Agents Are Ordering From DoorDash, and the Early Numbers Show It
Ars Technica and TechCrunch report that AI agents are barely placing DoorDash orders, while the company says it is open to working with outside agent makers.
Ars Technica reports that almost no one is using AI agents to place food orders on DoorDash. The outlet describes the practice as still in its early days, a relatively new way of shopping online that has not yet found a wide audience.
TechCrunch, citing the company, says DoorDash has received only several hundred orders that came from AI agents. That figure is small when set against the volume of orders the platform handles through its own apps and website, though the reporting does not say how the company counted agent-placed orders or over what period.
According to TechCrunch, DoorDash also said it is open to working with third-party agent makers. That statement suggests the company is not closing the door on automated ordering, even as the current volume remains modest.
Taken together, the two reports paint a picture of a service that is technically reachable by agents but rarely used that way so far. Neither outlet's reporting, as summarized here, identifies which agent products were involved or how the orders were completed.
Background
An AI agent is software that can carry out a sequence of tasks on someone's behalf, rather than simply answering a question. In shopping, the idea is that a user states a goal, such as a dinner order or a grocery restock, and the agent searches options, chooses items, and completes checkout. Supporters see this as a possible next step in online retail, where software does the tedious comparison work and the person only approves the final result.
Food delivery is an appealing test case because orders are frequent, the choices are familiar, and the stakes of a mistake are relatively low. Delivery platforms, however, are built around their own customer apps, which handle location, menus, promotions, payment methods, and tipping. Letting outside software operate those flows raises practical questions about how accounts are authenticated, who is responsible when an order goes wrong, and how the platform keeps its pricing and fee structures intact.
Companies in this space have generally approached automation through partnerships and interfaces rather than letting arbitrary software scrape a website. Formal integrations let a platform set rules, monitor abuse, and share in the value of the transaction. They also give agent makers a stable way to connect, which matters if they want to build products that work across many services.
Consumer trust is another factor. Many people are comfortable letting a phone remember a payment card but are less comfortable letting a program make purchasing decisions without close supervision. Early adoption of any new buying method usually depends on whether the convenience outweighs the effort of setting up and checking the results.
Finally, the broader technology is still maturing. AI systems can make errors, misread instructions, or misunderstand preferences, and those risks become more visible when the system spends real money. Any retailer opening its service to agents will have to weigh convenience for users against those reliability concerns.
Why it matters
The reports matter because they offer a reality check on a popular idea. Discussion of AI agents often assumes that automated shopping is around the corner, yet the early numbers reported here suggest that demand is still limited, at least for food delivery. That gap between expectation and usage is worth noting for anyone building or investing in agent products.
For DoorDash, the stance is a hedge. Saying it is open to third-party agents signals that the company wants to be positioned if usage grows, without committing to a particular path. For the wider delivery market, the message is that agent-driven orders are a possibility to plan for rather than a current source of meaningful revenue.
The story also highlights an unresolved question for online commerce: who controls the customer relationship when software sits between a person and a store. If agents become common, platforms may need to decide whether they are gatekeepers, partners, or simply endpoints. Each choice affects data access, fees, and the ability to influence what gets ordered.
For ordinary users, the practical stakes are modest today, but the direction is significant. Automation that works well could reduce friction, while automation that works poorly could create mistaken charges or confusing receipts. How companies handle those edge cases will shape whether people are willing to hand over more of their purchasing.
What to watch
Watch whether DoorDash publishes clearer detail on how agent orders are placed, verified, and supported, and whether it names any formal partners. A defined program would be a stronger signal than a general openness to working together.
Look for how other delivery and grocery platforms respond. If several competitors open similar channels, agent ordering could move from novelty to a standard feature. If they remain quiet, the category may stay niche.
Keep an eye on usage figures in future reporting. The current numbers are small, and the reports as summarized here do not say how the figures were measured, so later disclosures will show whether growth is real.
Finally, watch for questions about accountability, such as refunds, disputed orders, and payment protections. Those details will determine whether consumers feel comfortable letting software place their orders.
Viewpoints
Commentary from our columns: three angles on this story.
The Skeptic open questions and weak points
The headline is easy to misread. Several hundred orders is a small number, and the reports do not explain how it was counted, over what period, or whether any of those orders were repeat purchases by the same tool. Before anyone calls this a trend, it is worth asking what counts as an agent order in the first place. Is it software that clicks through the checkout, a formal integration, or a user who pastes a list into a chatbot and then places the order by hand? Each answer leads to a very different conclusion. The company's openness to outside agent makers is also a statement of intent rather than evidence of demand. Plenty of services say they are ready for a new channel long before customers arrive. The honest reading is that the idea is still unproven in this category. Whether agents become a real ordering method will depend on details that the current reporting leaves open.
The Ledger the business angle
For a delivery platform, every order is a transaction with fees, promotions, and customer data attached. An agent that places orders is useful to the platform only if it brings in new demand or at least does not cannibalize existing sales. That is the business question underneath this story. Opening the door to third-party agent makers could create a new distribution channel, but it also raises questions about who owns the customer relationship. If an outside service sits between the diner and the app, it may want a share of the fees or control over the data. Platforms will weigh that against the risk of losing direct contact with their users. The small volume reported so far means the economics are still theoretical. Investors and operators should treat agent ordering as an option to monitor, not a revenue line to model yet. Early partnerships will reveal who expects to capture value.
The User what changes for people who use it
For most people, the real question is not whether an AI agent can place a food order but whether it saves time without creating headaches. A person who orders the same meal every week might appreciate a tool that handles it. Someone who likes browsing menus, checking reviews, or changing their mind at the last minute may find an agent more annoying than helpful. The practical details matter a great deal: whether the agent asks for confirmation, how it handles substitutions, what happens when a restaurant is closed, and how refunds work if something goes wrong. Those are the moments when trust is built or lost. The fact that very few people are using this yet suggests most have not found a compelling reason to switch. That may change if the experience becomes simple and predictable. For now, the sensible approach for users is to pay attention to what permissions they grant and to review each order before it goes through.