Meta's consumer bet on Muse asks users to trust an ad company with sensitive data

Muse starts by proposing its own work. The agent offers a bunch of suggestions for things it can do, and one of the first things TechCrunch's Sean O'Kane did with it was act on one of them ^[7]. On TechCrunch's Equity podcast ^[21], Kirsten Korosec asked whether he saw Muse succeeding in its bid to become part of every part of his life. He said, "I mean, not really." ^[18]

At Connect, Mark Zuckerberg made it clear the company plans to push AI features everywhere ^[1]. Muse also inherits the promise of OpenClaw, which O'Kane called the first big explosion of people thinking, "Holy smokes, these agents can do all this stuff for me while I'm out and about, and I can just text with it and let it control my whole computer." ^[17]

O'Kane used an app that works like a relatively good chatbot, and he said it "does seem pretty powerful" ^[6]. The agent found him some unclaimed money ^[8]. His doubt was about ongoing use ^[9].

The discussion did not include usage or retention figures for Muse, so one reporter's couple of weeks with the app ^[4] is the only user report in it. The attention, at least, is documented. The hosts said Meta's AI announcements seemed to steal the spotlight during a week of new model launches from OpenAI and Anthropic ^[19].

The strategic case holds up better than the product case so far. Anthony Ha said the enterprise push at OpenAI and Anthropic comes from costs and valuations so high that they have to make money at a scale he called essentially unprecedented ^[12]. His first reaction to Meta heading the other way was "Wait, did they not get the memo?" ^[13] His more charitable reading was that the rivals' move leaves Meta more room on the consumer side ^[20]. Korosec pointed to Facebook, Instagram and WhatsApp ^[15] and said Meta is very good at embedding itself in everyday people's lives and has an established track record of doing so ^[14].

That track record was built on feeds and messages. An agent that runs errands needs to know more about the person it works for, and the hosts raised whether users can trust Meta's AI with sensitive information ^[16]. O'Kane put the problem in terms of the business model. "Meta's business is to sell you ads," he said ^[10]. "And yes, they'll make the argument that the more they know about you, the more accurate and interesting the ads will be," he said, before adding, "wake me up when we get to that fever dream." ^[11]

I'd judge Muse, or any consumer agent a team is weighing, on two questions for each task. The first is whether the task recurs weekly or happens once. The second is whether it needs information a person would hesitate to give an ad-funded company. A one-off task with nothing sensitive in it is a demo, and O'Kane's scan sits in the one-off column by his own account ^[9]. A sensitive task that happens once tends to get tried for a concrete payoff. Low-sensitivity jobs that recur make a small habit. The recurring, sensitive box is the only one where a personal agent justifies ongoing access, and it is where O'Kane's objection applies most. I would count only that box as evidence Meta's consumer bet is working. The cost of that rule is a smaller, slower number. Session counts will look better, and they will mostly be measuring the demo box.