DoorDash has put a number on the AI agent commerce hype, and it’s small. On its third-quarter earnings call, the company said third-party AI agents like Muse, Instinct and Fo have sent it only “several hundred” orders in total. TechCrunch AI reports that this compares to the roughly 3 million deliveries DoorDash completes on an average day, according to its 2025 report.
If you’re building or investing around agentic commerce, this is one of the clearest real-world checks we’ve had so far.
📌 The Short Version
- The number: Several hundred orders from third-party AI agents, total. Not per day, not per week.
- The scale: That’s a tiny fraction of a single day’s volume at DoorDash.
- The stance: DoorDash isn’t fighting agents. It’s watching closely, building its own, and looking at a future ad business inside them.
- The takeaway: Agent-driven ordering is real but early. Most consumers still open the app.
🗣️ What Tony Xu Said
CEO Tony Xu answered an analyst’s question about agents with a line worth remembering: “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.”
He followed with the numbers. “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 said.
Xu added that users still place most orders through DoorDash’s own app. Even so, the company wants to learn more about agent use cases and build a healthy relationship with the companies making them. He also pointed to a possible future business: advertising inside those agents.
🤖 DoorDash Is Building Its Own Agents Anyway
The small third-party numbers haven’t stopped DoorDash from shipping its own agent tools. Over the past few months it has launched:
- Ask DoorDash (June): a chatbot that takes orders from prompts and photos
- A CLI agent (July): ordering from the command line, aimed at developers
- A text-to-order agent (September): order by sending a text message
- An MCP server: lets other chatbots place orders, focused on corporate ordering
Quick note on MCP: the Model Context Protocol is an open standard that lets AI assistants connect to outside services in a structured way. Offering one means DoorDash decides how agents get in, instead of letting them scrape its site or click through it.
🧭 Why This Matters
What stands out to me is the gap between the story and the data. For the past year, “agents will do your shopping” has been one of the loudest pitches in AI. DoorDash runs one of the biggest transaction platforms in the U.S., and the agent traffic barely shows up.
That doesn’t mean the idea is dead. It means a few hard problems are still unsolved:
- Trust: People don’t yet hand a bot their card and their dinner order without checking.
- Access: Platforms haven’t agreed on whether agents can get in at all.
- Habit: Tapping a familiar app is still faster than explaining your order to a chatbot.
The access question is splitting the industry. TechCrunch AI notes that Amazon has been blocking personal shopping agents, while Walmart has rolled out the red carpet for them, hoping to pull business from competitors. DoorDash sits in the middle. It’s open to agents, but it’s steering them toward its own front doors.
🔮 What to Watch Next
- Ad inventory inside agents. If DoorDash starts selling placements in third-party assistants, that tells you it expects real volume.
- Official integrations vs. scraping. Expect more platforms to offer MCP servers or APIs as the approved way in, and to block everything else.
- Earnings call disclosures. “Several hundred” is a baseline. If the number jumps by orders of magnitude in the next few quarters, the hype will have caught up with reality.
For builders, the message is practical. Agent commerce is a long game, and the platforms that control the transaction are setting the rules. Building on official channels beats trying to work around them.
The full report, including more details from the earnings call, is available at TechCrunch AI.