The market value of Meta’s shares have gained close to $200 billion due to the company’s confidence in the scalability of its new AI agent, Muse. However, certain Muse phone calls are being handled by contractors using a “human concierge” feature, according to early internal experiments. This is at the core of the investment thesis, but Wall Street has mostly dismissed it as standard product development. Investing in the service. Investing in the service.
A Launch that changed the Narrative
After making its launch swiftly rose to the top of the US app charts. On behalf of a user, it can send emails, make trip arrangements, purchase online, complete forms, and contact companies. It crossed 2.5 million downloads in around two weeks, and Meta’s shares rose 11% in a single session.
Shareholders were aware of the significant AI expenditure prior to the launch, but they found it difficult to identify a consumer product that would justify it. Ads, recommendations, and interaction had all improved thanks to AI, but those benefits were integrated into already-existing processes. Muse presented something separate, with noticeable adoption and a possible subscription model. With over 3.6 billion daily users on Facebook, Instagram, WhatsApp, and Messenger, Meta is apparently thinking about $20 to $100 monthly subscriptions. Even a modest conversion rate could bring in billions in recurring revenue.
The timing also helped. Revenue increased by 28% to $60.8 billion in the second quarter, but the operating margin slipped to 31%. Meta invested $31.1 billion on capital investment in the quarter and expects $130 billion to $145 billion in 2026 for servers, data centers and networks. Given the strength of the company’s main business and the recovery of tech equities, Muse shouldn’t receive all the credit for the gain. However, it offered investors a chance to offset the expenditure with revenue.
Meta’s distribution advantage is real. Muse may be advertised on Facebook, Instagram, and WhatsApp without having to pay for each new user. However, that only clarifies how users are reached by Meta. It doesn’t specify how much it will cost to complete the tasks they ask for.
The Cost Hiding behind the Hype
Reuters reported that Meta is testing a human-concierge feature in which contractors handle some Muse calls. Half of Meta’s employees have the option to opt out of the experiment, which the firm claims is intended to gather information on privacy and safety prior to a public release. It is not suspect in and of itself to use people to modify an incomplete product, and the optimistic prediction that human involvement will decrease as the system learns may come to pass. Investors simply don’t have the info.
Simple tasks like composing an email or locating a restaurant ought to be heavily automated. When a user requests that Muse contest a hotel charge, rebook a vacation they missed, or cancel a service whose provider would want to keep the consumer, the economics become more complicated. Negotiation, insufficient information, and businesses with little incentive to work together are all part of these duties. If the agent purchases the incorrect item, schedules the incorrect date, or discloses personal information, someone must also take responsibility.
An AI agent can incur actual costs without malicious intent. All it requires is authorization to take action and a misinterpretation of the request. Repeated model calls, connections to several external services, and an individual to handle exceptions may all be necessary for a single activity. Since the most frequent users will probably be $100 monthly subscribers, Meta’s most valuable clients may also be the most costly to serve.
There are merchants. Shopify has embraced Muse, adding Shop Pay and treating the agent as a new sales channel. Due to worries about illegal activities, privacy, and user experience, Amazon has prohibited Muse from making purchases on their website. Meta cannot force the commercial internet to adopt its agent, and the eventual economics will depend on agreements, technical integrations and prospective revenue-sharing relationships. None of that is captured in a download chart.
Strong Position, One number that Matters
None of this makes Muse a failure. Meta certainly has the best starting position in consumer AI because it is now part of daily lives for billions of people. WhatsApp provides Muse with a comfortable home and is aware of their communication style, the brands they follow, and the things that keep them interested. A start-up could not afford to pay for contractors, infrastructure, and early blunders over a lengthy learning period, but its advertising business can. Contractors may even help by revealing which requests confuse the software and where users feel uneasy.
The strategic payoff could go beyond subscriptions. Meta gains a better understanding of commercial intent if Muse becomes the location where customers initiate purchases, plan travel, or get in touch with businesses.This may boost its advertising business and give it control over transactions that currently start with Google, Amazon, or the app of a store.
For a time, the tale will be sustained by downloads, and paid conversions and retention will be important. The cost of a task that is successfully finished, however, could say more than any of the other figures. It would show the cost of computation, the percentage of tasks completed without human assistance, and the cost of error correction. A falling cost would show that Muse is learning and that subscription revenue can scale. An economy more akin to a concierge business than a software platform would be indicated by a persistently high one.
Investors have handed Meta approximately $200 billion before seeing such numbers. The gain can hold if automation rises, human intervention falls and Muse becomes a profitable gateway to online activity. It won’t be determined by the number of users who download the software, but rather by the money Meta receives each time Muse finishes a task.
