Why the First AI Gadgets Failed, and Why the Next Assistant Must Live On-Device
Tony Fadell says the first wave of standalone AI hardware is over.
Speaking on stage at the inaugural MIT Future Fest, he named the Rabbit R1, Humane AI Pin and Limitless pendant as "Gen 1" products that have been discontinued, according to reporting published on October 7, 2026 TechCrunch. Fadell is credited as the father of the iPod, a co-creator of the iPhone, and the founder of Nest, which Google acquired.
His verdict was direct. The early devices, he said, "did not meet any kind of need" and were "just interesting technology for geeks." They shipped and demos circulated, but no lasting daily use formed around them.
Fadell said the companies behind those devices called him to ask for help and he declined. He did not detail what advice was sought.
What should replace them is not a better pin or pendant. It is an AI agent that lives on-device. Fadell predicted a successful agent will have to operate on-device only for privacy and to stay lightweight. A cloud request sends data to distant servers, which adds delay, cost and exposure. Local processing keeps personal context on the device, but is limited by memory, power and model size.
He said "trust and safety will be paramount for anything turned over to AI intelligence." He tied that to how rare delegated help remains, saying "less than 0.01% of the world population has ever had a human assistant do anything."
On who could build that trusted agent, Fadell pointed to incumbents with chips and distribution. He said Apple is the only company, maybe one other, he could currently see delivering it. Apple, he noted, "has the hardware, chips and pieces but does not have all the AI."
That follows earlier criticism. In an interview reported by AppleInsider in January, Fadell criticized Apple Intelligence and said Apple got its car project wrong and should have kept going with the car AppleInsider. In July 2025 he had said Apple needs to get bold on AI Bloomberg.
The broader context here is less about design than about permission. Standalone gadgets asked users to adopt new hardware, new controls and new cloud data flows at once, without a clear job phones could not do. An on-device agent reverses that burden. It must earn delegation inside phones people already carry and lock with biometrics, passkeys and sandboxing.
For builders, worth flagging is the trade-off Fadell leaves unresolved. On-device only simplifies privacy and cuts network dependence. It also limits access to the largest models, fresh retrieval and cross-device work. The fix is clear partitioning of what runs locally, what never leaves the device, and what users explicitly allow to escalate. If that can be made legible, the payoff is broad. Personal assistance stops being a luxury and becomes infrastructure.
