The Maze: Personal AI agents have attracted millions of users, but many businesses still report few purchases through them. WIRED’s October 31 reporting across roughly 50 companies exposes an early measurement problem: Meta’s US and Canadian Muse setup figures and merchants’ order counts describe different things. Neither establishes the size of the new sales channel.
Source: Meta. A frame from Meta’s September launch demonstration shows Muse recommending a stroller and displaying ordering messages; it does not measure actual merchant sales.
The Details:
The merchant evidence is modest, and uneven. Food-delivery service DoorDash described orders from Muse and other agents as numbering in the hundreds. Hotel operator Hilton called agent bookings insignificant. Online marketplace eBay described activity as very small. These are companies’ observations, gathered through earnings calls and interviews, rather than a representative survey. They do not share a common reporting period or a published method for identifying agent orders. The evidence challenges a sweeping disruption story, but cannot establish that personal agents generate few purchases everywhere. Meta also pointed WIRED to retail partners whose latest earnings reports predated Muse’s launch, leaving part of the picture unpublished.
Meta measures a different part of shopping. On its October 21 earnings call, Meta said more than 10 million people in the US and Canada had set up Muse. Zuckerberg said more than a quarter of people using it engaged in shopping, with average purchases above US$200. Those statements do not disclose total completed orders or allow a merchant conversion rate to be calculated. Muse is software that can browse sites and fill forms for a person. Meta says it seeks approval before buying and can pay through a wallet that supplies a one-use card. Finding a product, approving payment and completing an order remain separate steps.
Ordinary checkout rules can still trip up the agent. Restaurant chain Chipotle described problems applying loyalty discounts and ordering outside opening hours. Its digital chief said Muse’s debut increased agent orders by a single-digit percentage; the underlying volume was not disclosed. Chipotle was adjusting its website code. Other businesses are building approved routes instead of leaving agents to navigate ordinary pages. Insurance comparison marketplace Insurify reopened access on October 28 through connections and a dedicated interface for agents. It requires consent, preserved coverage information and identified access, with customers proceeding to the insurer to complete the policy. That is a controlled transaction process, not simply a better product search.
MarketMaze Take: Merchants can prepare for agents without pretending account registrations are sales forecasts. The useful early evidence is where an authorized purchase succeeds or fails. Chipotle’s discount problems and Insurify’s controlled reopening make that task concrete; neither tells us how large the eventual channel will become.
Reliability earns its budget before scale does. For merchants outside the initial US and Canadian Muse market too, the portable lesson is to measure completed, authorized orders and fix observed failures. A working discount or accurate insurance quote is a clearer investment case than a share of millions of agent setups. The interview sample cannot establish worldwide demand.
The future fee needs future proof. Meta expects businesses to pay a fee on transactions Muse facilitates, but the inspected call gives no rate or start date. A new channel might bring incremental customers while adding another intermediary’s charge. Before assigning it a growth budget, a merchant needs its own completed-order evidence and the eventual commercial terms.
Sources: WIRED, Meta, Meta Newsroom, Insurify
