DoorDash Delivered 993 Million Orders Last Quarter. AI Agents Placed “Several Hundred.” A Reality Check on Agent Shopping

By Thorsten Meyer | October 30, 2026 | ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on tech for your team

• Business-only prices and quantity discounts
• Tax-exempt purchasing
• Multiple users, one account, clear invoices

Create a free Business account →Free to register

As an affiliate, we earn on qualifying purchases.Hide

For two months, the tech world has talked as if AI agents were about to take over shopping. Meta says more than 10 million people set up its Muse agent in six weeks, and that shopping is the most common thing they do with it. Walmart built a special entrance for agents. Amazon locked its doors against them.

Then, on Thursday morning, DoorDash put a number on it. Asked on its earnings call how much food and grocery ordering outside AI agents now handle, CEO Tony Xu said: “we’re seeing maybe several hundred orders in total from some of these third-party agents.” In the same quarter, DoorDash counted 993 million orders.

The same week, Amazon and Visa said much the same thing in their own words. Here is what was actually said, why agents still rarely buy, and what that means for companies and for you.

Key takeaways

• The number: DoorDash’s CEO said outside AI agents have generated “maybe several hundred orders in total” since they started, a “fairly low use case”. DoorDash handled 993 million orders in the third quarter of 2026 alone.
• The scale: Even if “several hundred” meant a full 1,000, that is about one order in a million of a single quarter (my calculation).
• Not just DoorDash: Amazon’s CEO said on the same day that “few customers are doing much shopping this way yet”. Visa’s CEO spoke of an “albeit limited number of agentic transactions” and called trust “the biggest barrier to scale”.
• The other side: Meta says shopping is Muse’s most common use, with an average purchase above $200. These are Meta’s own, unaudited figures, and they say little about everyday orders like dinner.
• Why agents rarely order: the app is already fast, the hard part comes after the click, payment needs trust, many doors are closed or narrow, and early checkouts inside chatbots converted poorly.
• DoorDash’s bet: It calls this “atoms vs bits”. It builds its own agents but puts its money into the physical network: warehouses, autonomous delivery, and a $17 billion deal to buy Instacart.

Products Worth Considering

Amazon Kindle Paperwhite (12th Generation, 2024), 16 GB, 7-inch Display, Jade

AmazonView Latest Price

Amazon Kindle Scribe Colorsoft 64GB (1st Generation)

AmazonView Latest Price

Amazon Kindle Paperwhite Signature Edition (12th Generation, 2024), 32 GB, Metallic Black

AmazonView Latest Price

As an affiliate, we earn on qualifying purchases.

What DoorDash actually said

DoorDash published its third-quarter results on 29 October 2026 and held its call with analysts the same morning. The topic came up twice.

First, Deepak Mathivanan of Cantor Fitzgerald asked whether DoorDash wants to power food agents itself or leave some of that to “more generic agents like Muse”, Meta’s personal agent. Xu answered:

“I would say that this is something that is still a very dynamic area. And my take on all things technology is you never want to fight gravity. But at the same time, you don’t want to be delusional. What I mean by that is if you look at where AI agents are today, we’re seeing maybe several hundred orders in total from some of these third-party agents.”

He added that for an easy task like ordering food, “the predominant use case of that is actually through our app.”

Later, Dominic Ball of Rothschild & Co Redburn asked how DoorDash’s advertising business holds up if people start shopping through agents, and whether DoorDash could share revenue with them. Xu repeated the figure: AI agents “are generating several hundred orders across the time that they’ve been up and running. So it’s been a fairly low use case.” He said he expects “advertising opportunities even within some of these AI agents”, but “right now, with the low volumes”, there “isn’t yet that value.”

Status: confirmed. The quotes come from the call transcript; TechCrunch first reported them on 30 October.

One in a million: putting “several hundred” in context

DoorDash’s own release gives the comparison. In the third quarter of 2026, total orders rose 28% from a year earlier to 993 million. That is about 10.8 million orders a day (my calculation, 92 days).

“Several hundred” is a spoken, rounded figure, and Xu said it covers the whole time these agents have been running, not one quarter. So the comparison below is generous to the agents:

• Assume a full 1,000 agent orders.
• Assume all of them landed in the third quarter.
• 1,000 out of 993 million is about 0.0001%, roughly one order in a million.

One in a million is one second out of eleven and a half days. That is where outside agents stand today at one of the world’s biggest delivery platforms.

Same week, same message

DoorDash was not alone. Within nine days, four large companies described agent shopping on their earnings calls, and three of them were cautious:

• Meta, 21 October: Mark Zuckerberg said “more than 10 million people in the US and Canada have set up Muse personal agents” in about six weeks. “The most common use case for Muse is shopping,” he said, with more than a quarter of users engaging in it and “an average purchase size over $200”. In the first week, Meta said, it had to work with Stripe to raise credit limits because volumes were higher than expected. These are Meta’s own figures; Meta gave no total sales value.
• Visa, 27 October: CEO Ryan McInerney said that “in agentic, trust is the biggest barrier to scale” and spoke of the “albeit limited number of agentic transactions today”. Asked about Muse and another personal agent, Instinct, he said “it’s still very early”, and in Visa’s overall business “it’s a very small amount”.
• Amazon, 29 October: Andy Jassy asked his own question: does shopping through outside “horizontal agents” hurt Amazon’s ad business? “The answer is no. Few customers are doing much shopping this way yet as the experience is still not great nor more convenient than coming to a retailer directly to research or one-click order there.”
• DoorDash, 29 October: “maybe several hundred orders in total.”

Note what Amazon counts instead. Its own shopping assistant, Alexa for Shopping, was used by more than 400 million customers in the last twelve months, Amazon says, and customer interactions rose 570%. Amazon wants that assistant, not an outside agent, to be “the primary shopping agent customers use to shop Amazon”. We covered Amazon’s quarter in AWS Is Growing at Its Fastest Pace in Five Years, and the fight between Amazon’s closed door and Walmart’s open one in Amazon Blocks AI Shopping Agents, Walmart Lets Them In.

Hype and usage: how both can be true

Meta’s numbers and DoorDash’s numbers look like they contradict each other. They probably don’t. They measure different things.

Meta counts people who set up Muse and the share of users who shop with it. DoorDash counts orders that reached it from outside agents. An agent can be popular and still almost never order dinner. People seem to use agents where a purchase takes thought: a jacket, a trip, a gift. Xu said as much: he sees room for “long consideration purchases” and for purchases “proactively made on their behalf”. For dinner, people open the app.

There is also a counting question. Walmart described it plainly on 21 October: today, when an agent shops, “it signs in using a customer’s account”, so its purchases “appear together, without any differentiation.” If an agent logs in as you, the shop may simply see you. DoorDash did not say how it identifies agent orders. Its figure may miss some. But even ten times “several hundred” would still be a rounding error next to 993 million.

Why agents still rarely order

Put the statements side by side and five reasons stand out.

1. The app is already fast

Ordering your usual from a delivery app takes a few taps. Your address, your card and your favourites are saved. An agent adds a step: you explain what you want, then you check what it did. Jassy’s phrase fits: not “more convenient than” ordering directly. For a task you do every week, a helper only helps if it saves time.

2. The click is the easy part

Xu made the point that matters most for delivery: ordering food “is actually not the hard part”. The hard part is “making sure that they actually show up”, in the order and the condition you wanted, “especially when things go awry”. A missing drink, a melted ice cream, a substitute you didn’t want: someone has to fix that. An outside agent can place an order, but the restaurant, the courier and the refund still run through the platform.

3. Paying needs trust

Letting software spend your money is a big step. Visa’s CEO called trust “the biggest barrier to scale”. The industry is building safeguards: Walmart’s new agent profiles let you approve every purchase or set a spending limit, for example $50, $100 or $200, and agents “don’t receive customers’ payment-card information”. Visa is building special payment tokens for agents. Those tools are only now arriving.

4. Many doors are closed or narrow

Amazon blocks personal agents such as Muse, as TechCrunch notes. Walmart is opening an “Agent Gateway” and agent profiles that are “launching in the next month”. DoorDash’s official doors for outside agents are narrow. Its ordering connector uses MCP, the Model Context Protocol, a common standard that lets AI tools talk to outside services. It went into a limited beta on 30 September, only for companies ordering for their own employees, with early partners such as Vercel, Cognition and Mercor. A command-line tool is in beta for developers who build agents for personal use. Without such a door, agents have to use websites built for people. Walmart calls that “slow and error-prone”. DoorDash did not say which agents its several hundred orders came from or how they connected.

5. Checkout inside a chatbot disappointed

This was tested. From November 2025, Walmart sold about 200,000 products through OpenAI’s Instant Checkout inside ChatGPT. In March 2026, Daniel Danker, Walmart’s executive vice president of product and design, said purchases made inside the chat converted at about one third of the rate of shoppers who clicked through to Walmart’s own site, and called the experience “unsatisfying”, as MarTech reported. OpenAI began phasing out in-chat checkout in favour of the merchants’ own checkouts. Walmart now brings its own assistant, Sparky, into ChatGPT, Gemini and Muse instead.

Status: reasons 1 to 4 rest on the companies’ statements above; reason 5 on reporting from March 2026. The ranking of the reasons is my reading.

DoorDash’s real bet: atoms, not bits

DoorDash’s quarterly release has a section called “Atoms vs Bits”. Bits are the digital layer: apps, assistants, chatbots. Atoms are the physical world: couriers, warehouses, groceries, the bag at your door.

DoorDash writes that when ChatGPT launched four years ago, it saw two battles: “one to be the leading digital personal assistant, and the other to become the leading physical network to help fulfill tasks in the real world.” It chose the second. “As AI personal agents are now starting to become more mainstream, our strategy today remains focused on winning the battle for atoms.”

That does not mean DoorDash ignores AI. It has shipped its own agents: Ask DoorDash in the app, Text DoorDash for ordering by text message, a command-line tool for developers, and the MCP connector for companies. But the big money goes elsewhere: warehouses, autonomous delivery, and Instacart. On the same morning, DoorDash agreed to buy Instacart for $65 a share, an equity value of about $17 billion. The deal still needs Instacart’s shareholders and regulators and is expected to close at the end of 2027 or in the first half of 2028.

The logic is simple. Whoever wins the assistant on your phone still needs someone to bring the groceries. If agents ever do order dinner at scale, DoorDash wants to be the network they have to use, and it wants to keep its own app the fastest way to order until then.

What this means for companies

• Don’t plan on agent orders yet. For everyday purchases, the volume is close to zero at DoorDash and “few customers” at Amazon. Treat agent sales as an experiment, not a forecast.
• Count them separately. You can’t manage what you can’t see. Walmart’s agent profiles show one way: give each agent its own identity, so its orders, returns and complaints are visible.
• Be readable anyway. Agents already help people research and compare. Accurate product data, prices and availability that software can read are useful even if the final click happens in your own app.
• Choose your door on purpose. Open gateway (Walmart), walled garden (Amazon) or a narrow, controlled door (DoorDash’s beta). Each choice trades reach against control of the customer. Our 26 October analysis compares the options.
• Own the hard part. Delivery, substitutions, refunds and support are where customers decide whether to come back. DoorDash is spending billions on exactly that.
• Watch the ad question. Both DoorDash and Amazon tie the agent question to advertising. If recommendations become paid placements, trust may become the scarce good. See When the Answer Carries the Ad.

What this means for users

• For your usual orders, the app is still fastest. That is not a failure on your part. The companies say the same.
• Agents are better at research than at checkout. Comparing options, tracking prices, finding a gift: that is where they help today.
• If you let an agent buy, set limits. Approve every purchase or set a low spending cap where the shop offers it. Use one-time or virtual cards where available. Never hand an agent your bank password; our guide Don’t Give Your AI Agent Your Bank Password explains safer ways.
• Know who fixes mistakes. At Walmart, for example, the store’s returns process applies if an agent orders the wrong thing, and the agent’s own privacy policy covers what you share with it. Check both before you connect an account.
• Ask whether a tip is an ad. As agents start to carry advertising, a recommendation may be paid for.

Honest limits

• “Several hundred” is a rough, spoken figure. Xu gave no exact count, no list of agents and no definition of an agent order. The time window is “in total” since these agents started.
• Counting is hard. Agents that log in as the customer may look like ordinary orders, so the true number could be higher. It would still be tiny next to 993 million.
• One platform, one kind of purchase. Food and groceries are habit purchases with a fast app. Agents may do better with travel, electronics or gifts.
• Self-reported numbers. Meta’s, Amazon’s and Visa’s figures are the companies’ own and not audited. Amazon’s “50% more per order” for Alexa for Shopping users compares users with non-users; it doesn’t show that the assistant causes the extra spending.
• Interests. DoorDash wants you in its own app and sells advertising there. Its caution about outside agents is also strategy.
• It is early. Muse launched on 8 September. Walmart’s agent profiles are not live yet. These numbers can change quickly, and we will revisit them after the next round of earnings.

More on the companies building the plumbing of the agent economy: Four AI Deals in One Day. On what Meta’s agent can see: Don’t Use Muse? Meta’s AI Agent Can Still Read Your Messages.

Sources

• DoorDash: Third Quarter 2026 Financial Results, 29 Oct 2026 (total orders, Instacart agreement, “Atoms vs Bits”)
• DoorDash Q3 2026 earnings call, 29 Oct 2026: transcript via The Motley Fool (Tony Xu’s answers to Deepak Mathivanan and Dominic Ball)
• DoorDash: DoorDash brings office ordering to the AI tools teams already use, 30 Sep 2026
• TechCrunch: DoorDash said it’s only received ‘several hundred’ orders from AI agents, 30 Oct 2026
• Amazon: Third Quarter 2026 Results, 29 Oct 2026, and Q3 2026 earnings call, 29 Oct 2026 (transcript via Earnings Whispers)
• Visa: fiscal Q4 2026 earnings call, 27 Oct 2026 (transcript via The Motley Fool)
• Meta: Q3 2026 earnings call transcript, 21 Oct 2026, and Introducing Muse, 8 Sep 2026
• Walmart: Making It Possible for Agents to Shop on Behalf of Customers, 21 Oct 2026
• MarTech: Walmart says ChatGPT checkout converted 3x worse than its own website, 20 Mar 2026