Tony Fadell: first wave of AI devices failed because they didn't solve a real problem

Tony Fadell took the stage at the inaugural MIT Future Fest and put up three slides: the Rabbit R1, the Humane Ai Pin, the Limitless pendant. Each had generated serious hype. Each has already been discontinued. "Those were sort of 'Gen 1' AI products, and because of my background the companies came to me asking for help," Fadell said. "I didn't do it." The man behind the iPod, co-creator of the iPhone, and founder of Nest — acquired by Google — explained that none of those gadgets addressed a genuine need. They were interesting technology for enthusiasts that never actually fit into most people's lives.

The core problem, Fadell argued, is a fundamental mismatch. Less than 0.01% of the world's population has ever employed a human assistant. So when companies talk about a "personal assistant," they're speaking from the perspective of people who already have one. "Most consumers don't even know what an assistant is," he said. From his own experience, it took years to learn how to use an assistant properly, and then to trust that person with the most sensitive information — scheduling, banking. That trust doesn't appear overnight, and the path to the average consumer is still long.

Trust and safety have to come first. It's hard to talk about building trust without sounding detached, but the point stands: when you hire a human assistant, you don't hand them your bank credentials on day one. We're entering a world where everyone could have an AI assistant, but those tools are only valuable if they can be trusted completely. Right now, establishing trust in an AI assistant looks even harder than with a person. Meta recently launched Muse, a general-purpose AI assistant, but users have reason to be wary. A security researcher found a critical vulnerability shortly after launch, and a 404 Media investigation revealed that Meta employees discovered security issues that forced several teams into a "crazy race" to patch them before release, working overtime.

"Trust and safety will be critical in anything we hand over to any form of intelligence," Fadell said. "The only company I see right now, maybe one other, that can pull this off is Apple. They have all the hardware, all the chips, all the pieces — they just don't have all the AI stuff." Even if Apple isn't the one to ship the stickiest agent, Fadell predicts a successful agent will need to run entirely on-device — both for privacy and to keep the technology lightweight.

He's skeptical of the cloud narrative. "When you hear all the talk about 'data centers will conquer the world'... I'm not buying it," he said. "I've seen this before. I saw it play out in the early internet days. We have a ton of compute on device now, and more coming, and it's all still battery-powered." Keeping sensitive data on the device instead of routing it through the cloud is exactly how Apple earned user trust with biometric data via Face ID. In Fadell's view, Apple sees that model as the foundation for the future of personal AI agents.