Meta pushes consumer AI agent Muse at Connect while rivals chase enterprise

Meta used its annual Connect conference to put Muse, a new personal AI agent, at the centre of its plans, with chief executive Mark Zuckerberg signalling that AI features will be pushed across the company's products. According to TechCrunch, the announcement dominated a week that also brought new model releases from OpenAI and Anthropic, and it stood out because of the direction Meta chose.

A consumer bet in an enterprise race

As TechCrunch's Anthony Ha observed on the Equity podcast, most frontier AI labs have been drifting toward enterprise and coding tools, driven partly by enormous costs and valuations that demand revenue at an unprecedented scale. OpenAI and Anthropic, he noted, are planning to go public within roughly the next two years, which sharpens the pressure to monetise.

Meta went the other way. Alongside Muse, the company showed a cute, Tamagotchi-style AI companion device that it insists is intended for adults. Ha described the consumer pivot as head-spinning, but allowed that if OpenAI and Anthropic are converging on enterprise customers, the consumer side may be where Meta sees open ground.

Colleague Kirsten Korosec argued the strategy plays to Meta's strengths: Facebook, Instagram and WhatsApp have given the company a long track record of embedding itself in everyday life, and it has never really been an enterprise business anyway.

Hands-on: unclaimed money, then what?

TechCrunch's Sean O'Kane had already spent time with Muse, which had been available for a couple of weeks before Connect. He characterised it as roughly Meta's ground-up mobile version of OpenClaw, an agent product whose creators Meta bought and absorbed, wrapped in a chat interface that can take actions on a user's behalf.

His first notable result: Muse suggested checking whether any unclaimed funds existed in his name, and it found some. A cheque is on its way to him. But he called the win a one-off — "a party-trick type thing" — rather than something that drives repeat usage.

The more durable use case Meta has been promoting, according to O'Kane, involves deeper access: linking a credit card or a Gmail account so the agent can cancel unused subscriptions and flag duplicate charges, similar to services like Rocket Money and tasks he argued credit card companies should already handle. That is precisely where he hit what he described as the trust wall.

The trust problem

Somewhat to his surprise, O'Kane said Muse did not immediately plug into his Instagram, Facebook or Threads accounts on setup; it initially treated him like a stranger, which made him more willing to experiment. Over time, however, the app pushes to pull those services into the system so it can learn more about the user.

His scepticism is rooted in incentives. "Meta's business is to sell you ads," he said, adding that the company's argument — that knowing more about you makes ads more accurate and interesting — does not persuade him.

He contrasted this with Apple's revamped Siri on the latest iOS, which he said can now perform surprising and helpful actions on his phone. He said he is far more willing to share sensitive information with Apple's assistant, both from a security standpoint and because Apple's business is not built on advertising.

Beyond the one-time money find, O'Kane said he had not encountered much else in Muse that felt genuinely useful, though he acknowledged its deliberately consumer-friendly tone is part of why people are talking about it.

Why it matters

Muse is a platform-level bet: Meta is wagering that the next phase of consumer AI is an always-available agent woven into apps that billions of people already use daily. If it works, features like subscription management could make AI tangible for people who never touch a developer tool, and it would let Meta monetise AI attention at a scale its rivals cannot match through app-store subscriptions alone.

But the launch also crystallises the industry's central tension. Agents only pay off when users hand over financial accounts, inboxes and personal context, and Meta's ad-funded history makes that ask harder for it than for competitors like Apple. Muse's success may depend less on what the agent can do than on whether users believe they can trust what Meta will do with what it learns.