Deliveroo's owner has had a few hundred orders placed by AI agents

"Maybe several hundred orders in total"

The name may mean little outside the US. DoorDash is the American delivery giant, and it also owns Deliveroo, a deal it completed on October 2, 2025. On Thursday, October 29, during its earnings call, Cantor Fitzgerald analyst Deepak Mathivanan used his second question on AI agents. Should DoorDash let general-purpose agents, like Muse, the personal assistant Meta launched in September, place orders for customers? Or should it build its own?

CEO and cofounder Tony Xu didn't dodge.

According to the Motley Fool transcript, he said DoorDash sees "maybe several hundred orders in total" from those third-party agents. Customers, meanwhile, still go through the app. Later, answering another analyst, he pinned down the counter: "several hundred orders across the time that they've been up and running." So it's a running total since the start, not a tally for the past three months. And Xu doesn't say which countries or which of the group's apps are involved: no figure is given for Deliveroo.

For months we've been told agents will stock the fridge while we sleep, and that a market worth hundreds of billions is coming. In May, we wrote about software rebuilt so your agent picks the app for you. Now a major platform has said, for the first time, how many orders those agents have actually sent it.

A few seconds of business

The other number that day gives the scale. The same October 29, DoorDash's earnings release reported 993 million Total Orders for the quarter ended in September, Deliveroo included. That's the company's own definition: all orders completed through its marketplaces and merchant services over the period.

Spread over time, that's about 125 orders every second, day and night, on average. Everything third-party agents have ordered since they showed up is something DoorDash handles in a few seconds. Roughly the time it took to read this sentence.

The two figures don't cover the same span: a running total on one side, three months on the other. If anything, that makes the gap wider. Xu put it in his own words: "the volumes are very low." And his "maybe" signals a ballpark, not a count.

The objection: the door only just opened

Before writing off agents, there's a serious caveat, flagged by Ars Technica: all of this is very new. DoorDash only launched its official agent connector, built on MCP (the standard that lets an AI plug into a service), on September 30. That's less than a month before the call. It's a limited beta, open only to companies that order lunch for their employees, with a handful of partners such as Vercel and Cognition. Muse, the agent the analyst mentioned, is itself seven weeks old.

Ars points out that Pizza Hut was tentatively trying online ordering in 1994, and more than half its orders now come through the internet. Judging agents on their first few weeks, the argument goes, is like judging the web on its first pizza site.

But the objection doesn't explain everything. That connector isn't built for consumer assistants: an agent like Muse has to take other routes, such as the web interface. And on the previous earnings call, on August 5, Xu was already saying, according to PYMNTS, that volume from partner agents was low. Agents were already knocking on the door. There just weren't many of them.

Is a pizza too easy for an agent?

The most interesting part is Xu's explanation. In his telling, people don't hand the easy stuff, like ordering food, to third-party assistants: they do that in the app. Agents might find a role in long, considered purchases, or in background work done for you before you've even asked.

A PYMNTS test, run the day after Muse launched, shows the problem. Three ordinary errands: toilet paper on Amazon, a Domino's pizza, a restaurant table. None went through, stalled by logins to share and apps to connect one by one. The same toilet-paper reorder done by hand on Amazon took under thirty seconds.

That's a test on three tasks, not a study. But it illustrates Xu's point: when the app is already in your pocket, the agent has to beat three taps on a screen.

The opposite bet exists, and it comes with numbers. In a forecast published on December 17, 2025, consultancy Bain estimated agentic commerce could be worth $300 billion to $500 billion in the US by 2030, or 15 to 25% of online shopping. And according to Bain, it's the simple, well-defined requests, like "the cheapest batteries, delivered tomorrow," that will move first.

So two opposite readings of the same task: Bain bets the simple stuff goes first, DoorDash's CEO that it stays in the app. The comparison has its limits. Bain is talking about all US online commerce, not food delivery, and it also counts agents that merchants run on their own sites. The same document notes that most consumers aren't yet ready to let an AI handle a purchase end to end.

DoorDash prefers its own agents

DoorDash has nothing against agents, as long as they're its own. Ask DoorDash, launched in June, builds a basket from a craving or a photo of a recipe. Text DoorDash, in waitlist beta since September 30, lets you order by message. The company doesn't say how many orders either one handles.

Those in-house agents aren't counted in the several hundred, which covers only agents from outside.

The number also has to be read alongside the interest of the person giving it. In its shareholder letter, DoorDash says it has picked its fight: not the digital personal assistant, but the physical network, what it calls the battle of atoms. A customer who orders through someone else's agent comes in by the side door: they no longer walk past the app's shopfront, or its ads.

Downplaying third-party agents suits DoorDash, then. The figure is still the company's own, but it was given out loud, with no published method, by a business that has every reason to want it small.

A promising market?

On paper, Bain says yes, with a figure in the hundreds of billions. In DoorDash's accounts, it amounts to a few seconds for now. The two don't quite contradict each other. The forecast is about 2030 and all of online commerce; the tally measures a habit that has barely started. What DoorDash's CEO suggests is that delivered dinner may not be the way in.

Xu summed up his stance in a line: in tech, you never want to fight gravity, but you don't want to be delusional either. He promises to follow customers if they want to order another way. For now, gravity is pulling them back to the app.