Uber and DoorDash Reveal AI Agents Drive Almost No Orders

AI agents accounted for about 500 Uber Eats orders and four ride-hailing trips during one of Uber's busiest weeks on record — a week in which the company completed 320 million trips overall. CFO Balaji Krishnamurthy disclosed the figures on Uber's third-quarter earnings call on Wednesday, offering the first hard measure of how little traction autonomous ordering tools have gained on major consumer platforms.

The numbers are stark. Those 500 agent-driven deliveries represent roughly one in every 640,000 trips Uber completed in the last week of October, when Halloween demand pushed volume to what Krishnamurthy called "one of our largest weeks." Four ride-hailing trips from agents amount to a rounding error against millions of daily rides.

"It is very, very early, and very limited traction that we have seen so far," Krishnamurthy said on the call.

What did DoorDash report?

The Uber disclosure came one week after DoorDash CEO Tony Xu told investors his company had received "several hundred orders in total" from AI agents such as Meta's Muse and Instinct. DoorDash delivers billions of orders annually, which places the agent share of volume well below one hundredth of one percent.

Both executives named the same class of tools: agents built by Meta and third-party developers that can browse apps and complete purchases on a user's behalf. Neither company reported a meaningful revenue contribution from the channel.

Why are shoppers holding back?

Users who have delegated purchasing to agents report a mixed record, according to prior Business Insider reporting. The tools have handled tasks such as buying plane tickets and groceries, and some shoppers value them for monitoring prices over time.

Two failure modes dominate the caution. Agents make ordering errors — wrong items, wrong quantities, wrong delivery windows. Users also worry about agents acting unpredictably with stored credit card details, which has pushed many to keep agents on a short leash rather than grant full purchasing autonomy.

The pattern echoes a familiar interface-design problem: users accept automation for low-stakes, reversible actions such as price tracking, but resist it for transactions that move money without a human confirmation step.

Are the platforms still building?

Yes. Despite negligible usage, both companies continue to ship agent-adjacent features of their own.

• DoorDash confirmed in September that it is testing a text-to-order feature that lets users place orders without opening the company's app, including simplified reordering of previous purchases.
• Uber continues to develop AI tools embedded in its own apps, according to comments on the earnings call.

The strategy suggests both companies view agent-mediated ordering as inevitable infrastructure rather than a present-day demand signal — building the rails now while measured adoption stays near zero.

What does this change for the agent ecosystem? It sets a quantitative baseline: as of late 2026, agent-initiated commerce on two of the largest ordering platforms in the world can be counted in hundreds of orders, not millions. Whether the next two earnings calls move that figure by an order of magnitude will indicate whether autonomous purchasing is a behavior waiting to form or a prediction outrunning the evidence.