DoorDash counted 'several hundred' AI-agent orders against 993 million — the first hard volume check on a $93 billion agentic-payments market

Asked on DoorDash's third-quarter earnings call on Oct 29, 2026, whether AI agents were creating incremental orders, chief executive Tony Xu said the company sees 'maybe several hundred orders in total' from third-party agents including Muse from Meta Platforms, Instinct and Fo — and that consumers still order 'predominantly through our app'. Against 993 million orders in the quarter, roughly 10.8 million a day across 92 days, even the top of 'several hundred' is fewer than one agent order per million. Puts third-party agent volume below 0.0001% of the order base.

Every rung on that ladder except the last is a projection of what the agentic stack will be worth, not a count of what it has already sold. Only one company has put a completed-order number on the record, and it did so because an analyst asked.

The size of the basket is the first constraint. DoorDash's average order carries about $34 of Marketplace GOV and $4.65 of revenue, and it repeats twice a week. That is among the least attractive things in the economy to delegate: the agent has to beat a two-tap reorder, and a human still has to hand over a payment credential. Visa's own agentic-commerce trust research, published Sept 9, 2026, found 72% of US consumers have already used an AI assistant, but only 23% trust generative AI to handle payment transactions on their behalf. Trust in the counterparty caps agent checkout volume — 61% of respondents said they would trust Visa to handle agent transactions.

That is the structural read: agents browse, humans buy. Visa's own framing is that agentic commerce is 'very early' and immaterial to results over the next four quarters, even with about 50 AI partners live in production and that number expected to double. Mastercard's agent stack — Agent Pay, Agent Connect, Verifiable Intent — ships for the same reason: to keep the transaction on the network when the buyer changes. Every rail is being built for a buyer who has not shown up yet, which is a defensive posture, not a growth plan.

The DoorDash disclosure exposes something about the rest of the stack that no balance sheet shows. Visa has shipped a trusted agent protocol, an Agentic Ready programme and roughly 50 live AI partners. Mastercard has Agent Pay, Agent Connect and Verifiable Intent. Shopify has Agentic Storefronts and a catalog layer built to be read by agents. Walmart has an Agent Gateway and Agent Profiles. [Stripe], which is private, has an Agentic Commerce Suite built on shared payment tokens. Not one of them publishes an agent-initiated order count or transaction volume.

The reason is not concealment; it is that there is no meaningful number to publish. Visa earned $12.14 billion of net revenue in its fiscal fourth quarter, up 13%, on payments volume up 9% in constant dollars and cross-border e-commerce volume up 17% — all of it conventional payment volume. Mastercard earned $9.81 billion, up 14%, on gross dollar volume of $3.0 trillion, up 7%, with value-added services up 19%. Both networks get paid on volume, not on who typed the order, so their agentic spending is cheap insurance against disintermediation rather than a bet on a new revenue line.

The platforms making real progress on agents are the ones that let agents in without giving the transaction away. Walmart announced on Oct 21, 2026 that it is building a Walmart Agent Gateway, a model-context-protocol route into Walmart and Sam's Club for agents, opening with a beta for companies and developers, with Agent Profiles to follow within a month so a shopper can authorise an agent, cap its spending and see its purchases separately. The acceptance test is price: the price an agent sees matches what a human sees on Walmart.com. Shopify is doing the same thing a layer up — Agentic Storefronts, live since March 2026, give merchants out-of-the-box access to ChatGPT, Copilot, AI Mode and the Gemini app.

For DoorDash the arithmetic is unusually forgiving. A third-party agent that places an order still lands inside the same 993-million-order marketplace, earns the same 13.6% net revenue margin and is invisible in the app. Its in-app assistant, text-ordering feature and Model Context Protocol keep the agent inside the platform. Third-party agent orders add nothing to the model and take nothing from it either, which is the opposite of what the agentic-commerce narrative assumes.

Two days before the earnings call, on Oct 28, 2026, DoorDash agreed to acquire Maplebear, the operator of Instacart, for $65.00 a share — roughly $17 billion in equity value, a 26% premium to the Oct 22 close, payable at least 75% in cash. The deal is expected to close at the end of 2027 or in the first half of 2028, and management frames it as a path to $500 million or more of annual net synergies by year three, with more than 100% of synergies reinvested in the first year after close.

Here is the causal chain that matters for investors. Third-party agent demand is not arriving, so a platform that owns checkout cannot wait for an agent economy to create orders; it buys the category where agents are most likely to transact and where the basket repeats. Morgan Stanley Research found in December 2025 that AI-driven purchases cluster in groceries and consumer packaged goods, and estimates agents could take 10% to 20% of US e-commerce — $190 billion to $385 billion — by 2030. Buying the grocery marketplace buys the catalog and the repeat basket that third-party agents still cannot build, which is why it is a better hedge against agentic commerce than agentic commerce itself.

Over one to three years, the risk to this view is that trust closes faster than the payment networks expect. If consumers begin to trust agent payments, DoorDash's own surfaces could pull a slice of that 993 million-order base into agent-initiated demand with no third party involved, and Meta Platforms's Muse would be pushing on the same door Amazon just closed. The larger risk is the one the evidence already supports: agent traffic stays a permanent sliver while capital keeps flowing to rails built to serve it.