Two. That’s the number of devices Meta needs you to carry for Muse to make sense — smart glasses on your face and a keychain-sized gadget called Muse Charm in your pocket. Not a chat window. Not a browser tab. Two pieces of hardware you have to actually want.
That framing is why Muse took the spotlight this week from OpenAI and Anthropic, both of which were mid model-drop. Meta didn’t win on benchmarks. It won on a category nobody else had shipped into yet.
What actually happened
Meta launched Muse, a personal AI agent, and reporting says it surpassed early metrics from competitors including ChatGPT. Meta Connect 2026 tied Muse directly to hardware — AR glasses first, extending into the Charm keychain device. Meanwhile Anthropic and OpenAI were rolling out models, and Anthropic CEO Dario Amodei published an essay on AI safety fears that made its own news cycle.
So the scoreboard for the week reads: two labs shipped models, one company shipped a product people can hold. Guess which one dominated coverage.
The part that should make you suspicious
“Surpassed early metrics of ChatGPT” is doing a lot of work in that sentence, and I want to be honest about what I don’t know. Early metrics for a Meta product means distribution through apps that billions of people already have installed. ChatGPT had to be discovered. Muse gets to show up. Those are not comparable starting conditions, and anyone treating the comparison as a quality signal is reading a marketing chart as a review.
I also can’t tell you how good Muse is at the things agents are supposed to be good at. Coverage describes it as an impressively capable agent with some caveats, which is roughly what every agent launch gets described as in week one. I’ve tested enough of these to know the gap between demo-capable and Tuesday-afternoon-capable is where most of them die.
The question TechCrunch asked, and it’s the right one
One headline this week put it plainly — would you trust Meta with your entire digital life? That’s not a rhetorical jab. It’s the actual product requirement. An agent living on your glasses and a device clipped to your keys needs access to your calendar, your messages, your location, your field of view. The value proposition is proportional to the access. A weak agent with narrow permissions is useless. A useful one sees almost everything.
OpenAI and Anthropic ask for a fraction of that today. You paste in what you want them to see. Muse’s pitch is the opposite: stop pasting, let it watch. Whether that trade is acceptable depends entirely on how much you trust the company holding the camera, and Meta’s history here isn’t a blank page.
I’m not going to tell you the answer. I’m telling you that’s the review criterion, not token throughput.
Why Meta’s strategy still might work
Consumer integration is an unglamorous bet and historically a winning one. The best model rarely wins the market. The default does. If Muse is the thing that answers when you talk to glasses you already bought, it doesn’t need to beat Claude or GPT on reasoning. It needs to be there, be fast, and not embarrass you in public.
Meta also got to the consumer AI device market before OpenAI did, which matters more than it sounds. Hardware takes years. Supply chains, retail, returns, firmware updates. Being early to a physical category buys time that a competitor can’t simply spend money to recover.
Why it might not
Zuckerberg’s strategy is unproven, and coverage this week said so directly. A few specific risks I’d watch:
- Two-device dependency. Asking people to carry glasses and a keychain unit is friction dressed up as an ecosystem.
- Agent reliability in public. Voice-first agents fail loudly. Text agents fail privately.
- Permission fatigue. The more access Muse needs, the more chances users have to say no.
- Attention is not adoption. Stealing a news cycle from two labs mid-launch proves Meta’s PR works, not that Muse does.
My read
Muse is the most interesting AI product story of the week because it’s competing on a different axis than everyone else. OpenAI and Anthropic are arguing about frontier pace. Meta is arguing about what’s on your face. Those are not the same fight, and for a lot of ordinary users the second one decides which assistant they actually use.
I’d hold off on declaring a winner. Ship dates, real-world reliability, and the permissions dialog will tell us more than any launch-week metric. Right now Muse has the spotlight and an unanswered trust question, and those two things tend to resolve in the opposite direction from each other.
🕒 Published: