Tony Fadell says AI gadgets failed the everyday-use test

The missing everyday use

Fadell is often called the “father of the iPod.” He also helped create the iPhone and founded Nest, the smart-thermostat company later acquired by Google. That history helps explain why makers of new devices sought his advice.

His diagnosis of Rabbit R1, Humane Ai pin and Limitless is straightforward: they were interesting technologies for enthusiasts, but did not solve a problem people encountered in daily life. The early devices promised the convenience of a personal assistant and then failed to deliver it reliably.

There is a deeper mismatch, too. Most people have never had a human assistant, so they may not know what they would ask one to do—or what it would take to trust one. Fadell says fewer than 0.01% of the world’s population have ever had a human assistant. When companies claim consumers need an assistant, he argues, they may be speaking from the experience of people who already have one.

Even for Fadell, working effectively with an assistant took years. Trust came gradually: first learning how to work together, then sharing confidential information and delegating meetings and bank interactions. A personal AI assistant may eventually be available to everyone, but access alone will not make it usable.

Trust is the product

Giving an AI assistant authority may be harder than trusting a person. Fadell points to Meta’s recently released general-purpose AI assistant, Muse. A security researcher soon found a serious vulnerability, while a recent 404 Media report said some Meta employees had identified security problems before launch and several teams rushed to fix them, working overtime.

For any system entrusted with decisions, Fadell sees security and trust as prerequisites. He thinks Apple has an opportunity to build such a product, perhaps alongside one other company. Apple has devices, chips and other necessary components, but, in his assessment, not everything it needs in AI.

The company’s existing privacy reputation gives it a potential advantage. Fadell says consumers view Apple more favorably than competitors on privacy; people, for example, share biometric data through Face ID. But Apple still lacks a world-class AI model of its own: the new AI-powered Siri runs on versions of Google’s Gemini created specifically for it.

Fadell’s preferred direction is an agent that runs on the device, keeping private data local and the system lightweight. He rejects predictions that data centers will take over, saying similar forecasts appeared during the growth of the internet. Modern devices already have substantial computing power and still run on batteries.

Why the gadget makers need hardware

Fadell’s account also offers a reason Meta and OpenAI are moving into devices: unlike Apple, they do not have billions of devices in circulation. Without access to a smartphone’s sensors, they must ask separately for permission to collect video, audio or location data. Those requests can grow to 20 items or more.

His explanation is that companies are building screenless devices with the sensors they need, then connecting them to phones over Bluetooth or Wi-Fi and sending data online or over 5G. That may widen access to sensors, but it does not by itself answer the more difficult question: what should an assistant do, and how does it earn permission to do it?

I think this is the more consequential gap in the current wave of AI hardware. A new device can make data collection easier; it cannot shortcut the years of use and trust that Fadell says an assistant requires.

One shot, or several

Fadell has criticized Apple when he thought it was wrong, despite his role in creating the iPod and iPhone. Asked about luck in finding product-market fit, he said startups have a harder time: a product failure can sink the company, leaving it with one shot. Apple, he joked, can afford more attempts—Vision Pro, for example.

That difference matters for AI gadgets. The companies building them may have room to experiment, but the products still have to earn a place in ordinary life. The next test is not whether an assistant can be put in a device; it is whether people will trust it enough to keep using it.