Tony Fadell: Gen 1 AI gadgets failed, on-device agents next

Fadell's slide at MIT Future Fest put three discontinued AI devices side by side — the Rabbit R1, the Humane Ai pin, and the Limitless pendant — and he noted that these companies had called him for help because of his background. He declined. His record includes the iPod, the iPhone, and Nest, the smart thermostat company acquired by Google, which is the basis for the authority he claims on the question of what makes a consumer device stick.

The thread running through his argument is trust. He compared an AI assistant to a human assistant, noting it took him a couple of years to understand how to use one and then to trust them with sensitive data and let them work with his bank. That is the gap he sees between the industry's vision of the future and where general consumers and businesses actually are. He applies the same test to launch-day security: Meta recently launched Muse, its all-purpose AI assistant, and a security researcher quickly found a serious vulnerability in it, while a 404 Media report states some Meta employees found security issues that prompted multiple teams to work overtime on a "mad dash" to fix them before launch.

Fadell's other prediction is architectural. He rejects the idea that data centers will conquer the world, pointing to how much power and compute now sits on battery-operated devices, and argues a successful agent will have to operate on-device only for privacy and weight. He sees one company, maybe two, positioned to do that, and says Apple has the hardware, chips, and pieces but not the AI — its new Siri AI runs on custom-built versions of Google's Gemini. Whether that gap closes, or a rival pairs a model with its own device, is likely to determine which assistant consumers end up trusting.