Tony Fadell told the audience at MIT Future Fest that the first wave of AI gadgets failed because none of them answered a real need. TechCrunch’s Amanda Silberling reported the stage talk on 7 October, and what follows is Fadell’s argument, not an established finding.
Fadell opened with a slide of three discontinued devices: the Rabbit R1, the Humane Ai pin, and the Limitless pendant. He said the makers of those products phoned him for help. “I didn’t,” he said. His verdict on the group was blunt: it was “just interesting technology for geeks.” He speaks with some authority on hardware. TechCrunch describes him as the “father of the iPod,” who helped build the iPhone, and he started Nest, the thermostat maker Google bought.
His deeper complaint is that the whole pitch rests on an experience almost nobody has had. “It’s less than 0.01% of the world population that has ever had a human [assistant] do anything,” he said, so a company promising an assistant is mostly talking to people who already employ one. Fadell described years of learning how to use a human assistant before he trusted that person with sensitive data and with errands that touch his bank. A consumer, in his telling, will need the same ramp with software.
Trust is where Silberling brings in Meta’s new Muse assistant. She writes that after launch one security researcher turned up a serious flaw in it, and she cites a 404 Media report that some Meta employees spotted security problems that sent several teams into a “mad dash” of overtime fixes before release. Those are TechCrunch’s and 404 Media’s reports, not Fadell’s claims, and the article describes no breach or exposed user data.
On Apple, Fadell makes a two-part argument. “Trust and safety are going to be paramount with anything that we turn over to some kind of intelligence,” he said, and Apple, perhaps with one other company, is the only one he can see doing it: “Apple has all the hardware, they have all the chips, they have all the pieces, but they don’t have all the AI stuff.” Both halves matter to his case. Silberling adds that Apple has no world-class model of its own, and that its new Siri is powered by tailored, custom versions of Google’s Gemini.
He also predicts that any agent that sticks will run on the device only, for privacy and to stay light. He dismissed the idea that data centers will conquer everything: “I don’t believe it.” Phones now carry plenty of compute, he said, though they still run on batteries. His theory for why Meta and OpenAI are building gadgets is that they lack the sensors already in your pocket: “they don’t have access to the sensors on your phone.” A standalone device lets them collect that data without asking the phone for it one permission at a time.
Apple’s own assistant leans on Google’s model, which suggests that hardware and goodwill have not yet closed the gap in model quality. An agent that books meetings and talks to a bank also has to reach outside the device, so “on-device only” may be a goal that shrinks the product more than it protects the user. The trust point holds up against that objection, while the architecture point reads as a bet.
The part of his talk that transfers to anyone shipping an agent now is the permission ladder. If your product asks for video, audio, location and account access in the first session, it is asking for what Fadell says even a seasoned executive took years to hand a human. Audit that first-run list before your next release.
Based on reporting by Amanda Silberling for TechCrunch, published 7 October 2026.
