Skip to main content
AI for Business
· 6 min read

DoorDash Got 'Several Hundred' AI Agent Orders. Muse Is Already Calling Businesses.

Flat-vector line-art illustration on a dark navy background: a vintage telephone at lower left in electric-blue outline with its coiled cord, a pointing hand entering from the right, and a dotted electric-blue arc rising from the phone to a small shopping bag on a slate-blue pedestal at upper right

Two numbers landed on the same morning that together tell you more about agentic commerce in 2026 than every keynote this year. The first, from DoorDash's third-quarter earnings call on October 29: after a year of "AI agents are coming for commerce," the company has received several hundred orders total from third-party agents like Muse, Instinct, and Fo — against roughly 3 million deliveries a day. The second, from The Globe and Mail's AI reporter: Meta's Muse personal agent, the app that topped download charts after its September launch, can now place live phone calls to U.S. businesses — and the stores that picked up mostly had no idea they were talking to software. Agent ordering is a rounding error. Agent phone calls have already started. For a local business, that gap is the whole story.

What DoorDash Actually Said

On the Q3 call, an analyst asked CEO Tony Xu about third-party agents ordering food. His answer, quoted by TechCrunch: "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." Several hundred. Not per day — in total. DoorDash's own 2025 reporting put its average at about 3 million deliveries daily, which means agent-placed orders are somewhere around one in a million of everything the platform moves. Customers still overwhelmingly order through the app, Xu said — while adding that DoorDash wants a "healthy relationship" with agent makers and can see a future advertising inside them.

The split verdict inside one quote is worth sitting with. DoorDash isn't dismissing agents — it built its own text-to-order agent in September, a CLI agent in July, and the Ask DoorDash chatbot in June, plus an MCP for corporate ordering. It's saying the demand isn't there yet, and it's not going to panic-build for it. Meanwhile the retailers are dividing: Amazon blocks personal agents outright, Walmart rolls out the red carpet to poach their business. Every business in between is being told to pick a side of a war that — by the only hard number anyone's published — barely has combatants yet.

The Calls, Though, Are Already Real

Same morning, different frontier. The Globe's Joe Castaldo tested Muse's new calling ability by directing his agent — which he'd named BigDawg — to phone U.S. music stores and ask whether a specific item was in stock and at what price. The calls worked. After each one the agent handed him a transcript. Two details from his account matter more than the novelty. First, BigDawg did not identify itself as AI; it said it was calling on a person's behalf and disclosed the recording, then "immediately barrelled into asking about products before anyone had time to process the weirdness of it all." Nobody questioned it. One store hung up. Second, it made real mistakes — one shop said it didn't have the item, and the agent started asking about pricing anyway.

Put that next to Yelp's Check for Me, which started sending AI agents to call restaurants about wait times last year, and the pattern is unambiguous: the consumer side of agentic commerce isn't arriving through shopping carts. It's arriving through your phone line, one call at a time, today — while the purchasing side is still a headline in search of volume. The banks see the same asymmetry: BMO, TD, and RBC have all moved to block third-party agents from customer accounts, and companies are publicly wrestling with "a tug of war under way for the customer experience" when a third-party app can recommend and buy without the customer ever seeing the brand.

Why This Matters for Your Business

If you've been feeling pressure to make your website "agent-ready" — agentic checkout, machine-readable catalogs, protocol integrations — DoorDash just handed you the permission to prioritize honestly. The largest food-delivery platform on earth, with every incentive to hype agent orders, says the total is several hundred. Your Temecula or Murrieta business does not lose revenue this quarter because an AI can't buy from your site. But Muse agents calling U.S. businesses right now — unidentified, mid-mistake, asking for prices — means the person (or software) asking your front desk "do you have this in stock and what does it cost" may be neither a customer nor honest about what it is. The speculative project and the live traffic are on opposite sides of your business, and most advice this year points at the wrong one.

There's a second-order effect worth naming. When an agent calls, the transcript is what the human actually reads — your answer becomes data, gets summarized, and steers a purchase you never see. The question stops being "how do I rank in AI answers" and becomes "what does my business say when software calls and asks."

The Reality-Check Playbook

  • Stop budgeting for agent checkout in 2026. The best number available says several hundred orders platform-wide against 3 million daily deliveries. Every hour spent on agentic payment protocols this year is an hour not spent on the channel where the traffic actually is. Revisit when a platform publishes a number that starts with a percentage, not "several hundred."
  • Prepare the phone for callers who aren't customers. The stock-check call is the most common agent errand today. Write the script your staff runs: item, yes or no; price, exact; anything you want the transcript to carry — hours, location, same-day availability. A clean answer to an agent is a clean answer to the human who reads its transcript.
  • Decide your disclosure posture for outbound AI calls. If you use an AI to call vendors or suppliers on your behalf, tell it to identify itself as an assistant — the standard you'd want inbound. Muse not identifying itself on its calls is the courtesy gap to not copy.
  • Keep the discovery layer honest. The reason agents call or browse at all is that a human asked one about your category. That's the same AI-referred discovery shift we covered when Instant Checkout retreated: agents decide who humans buy from long before they buy themselves. Your site's job is to be legible to a summarizer — clear services, clear prices, clear service area.
  • Watch one number, not the news cycle. Track how many of your inbound calls and form fills come from an AI assistant — just ask "are you calling for yourself, or did an assistant suggest this?" One line in your intake, zero new tooling, and in six months you'll have a better agentic-commerce dataset than DoorDash just published.

What a Temecula or Murrieta Business Should Do This Week

  • Script the three most common phone questions — availability, price, hours — so a nervous teenager, a temp, or an AI gets the same complete answer, transcript-friendly by default.
  • Make your site machine-legible where it's cheap: plain-language service list, prices or price ranges, service area, one obvious contact path. That covers the agent-summarizer discovery layer without touching payments.
  • Ask your phone answerer what happened yesterday. If anyone has already fielded an odd caller — fast, scripted, no small talk, recording disclosure — you've met the future. Note it.
  • Or make the phone self-defending. That's the job: PepeWebTech builds AI employees for Temecula and Murrieta businesses — an AI Chatbot at $397/mo for your site and messages, an AI Phone Agent at $697/mo that answers every call the moment it rings, or the Full AI Package at $997/mo — flat pricing, free setup, no contract, cancel anytime. Book the free demo and ask one question: when a customer's AI calls, what does ours say back? We'll show you the exact script.

The Bottom Line

A year of agentic-commerce hype produced, at the biggest delivery platform in the country, several hundred orders — a rounding error the CEO himself refuses to inflate. Meanwhile the same wave already reaches local businesses as phone calls from agents that don't announce themselves and sometimes ask the wrong questions. Build for the phone that's ringing, not the checkout that isn't. We track both sides of that line — the real numbers and the real calls — on the blog.

Sources

  • TechCrunch — DoorDash said it's only received 'several hundred' orders from AI agents (October 30, 2026) — the primary source for the earnings-call number: CEO Tony Xu on the Q3 call, "several hundred orders in total" from third-party agents like Muse, Instinct, and Fo, against a 2025-reported average of 3 million deliveries daily; "you never want to fight gravity. But at the same time, you don't want to be delusional"; DoorDash's own text-to-order agent (September), CLI agent (July), Ask DoorDash chatbot (June), and corporate-ordering MCP; Amazon blocking personal agents while Walmart welcomes them.
  • The Globe and Mail — Business Brief: Hello, it's your AI assistant calling (October 30, 2026) — the primary source for the calling detail: reporter Joe Castaldo's Muse agent ("BigDawg") placing calls to U.S. businesses to check stock and price, providing transcripts afterward, not identifying itself as AI while saying it called on his behalf and disclosing recording, no staff questioning it (one hang-up), and a confusion case where it asked pricing after being told the item was unavailable; Muse's September launch topping app-store charts; BMO, TD, and RBC blocking third-party agents from accounts; PwC's Danielle Gifford on tailoring services to agents.