Tony Fadell says the first wave of AI hardware failed for a simple reason: it didn’t solve a real problem. Speaking at the first MIT Future Fest, TechCrunch AI reports, the man credited as the “father of the iPod” put up a slide with three discontinued devices: the Rabbit R1, the Humane Ai Pin and the Limitless pendant. All three companies asked him for help. “I didn’t,” he said.
You don’t have to take that as a dig. It’s a diagnosis, and it tells us a lot about where personal AI goes over the next few years.
The Gen 1 Problem: Cool Tech, No Pain Point
Fadell’s main criticism is about product fit, not the hardware. “Each one of them didn’t meet any kind of need,” he said. “It was just interesting technology for geeks.”
He also has a sharper point about the “AI assistant” pitch itself. According to Fadell, fewer than 0.01% of people have ever had a human assistant. So the people pushing the idea of an assistant are mostly people who already have one. “Most consumers don’t even know what an assistant is,” he said.
He says it took him a couple of years to learn how to use a human assistant well. Only after that did he trust them with sensitive data, meetings and banking. AI agents are asking for that same level of trust from day one.
Trust Is the Real Bottleneck
What stands out to me is the timing. Meta just launched Muse, its all-purpose assistant. A security researcher found a serious vulnerability soon after launch. And a 404 Media report says Meta teams made a “mad dash” to fix security issues before release.
“Trust and safety are going to be paramount with anything that we turn over to some kind of intelligence,” Fadell said.
He thinks Apple, and maybe one other company, is best placed to win here. Apple has the hardware, the chips and years of privacy goodwill built up through things like Face ID. What it lacks is a top-tier model of its own. The new Siri runs on custom versions of Google’s Gemini.
Fadell’s Prediction: On-Device Beats the Data Center
Fadell’s boldest call is that the winning agent will run entirely on the device. That’s partly for privacy and partly to keep it lightweight.
“When you’re hearing all of these things about, ‘Oh, data centers are going to conquer the world,’ … I don’t believe it,” he said. He compares it to what he saw in the early internet days.
That cuts against hundreds of billions in current infrastructure spending, so give it the weight it deserves. Fadell has a strong record on consumer hardware: the iPod, the iPhone, and Nest, which Google bought. But he built devices, not models, and he’s clearly fond of Apple. He’s also not uncritical of it. He took a shot at the Vision Pro, noting that a startup “get[s] one shot” while Apple can absorb a flop.
Why Meta and OpenAI Want Their Own Devices
Fadell’s take on the new gadget push is the most useful idea in the talk. Companies without billions of phones in circulation can’t get at your sensors easily. Every app permission (camera, mic, GPS) is a separate ask, and the list grows past 20.
His read is that they build a screenless device that comes with all those sensors built in, then pipe the data through your phone or a 5G connection. If he’s right, the next round of AI hardware is less about form factor and more about getting access to data.
What Comes Next (2026-2028)
Here’s where I see this heading over the next one to three years:
- Trust will be built step by step. Agents that start with low-stakes tasks and earn more access over time will beat agents that ask for everything up front.
- On-device and hybrid setups will grow. Even if data centers don’t lose, privacy-sensitive features will move closer to the device.
- Security incidents will decide who wins. One major agent breach could set a whole company back years.
- Hardware startups face a high bar. If you only get one shot, you can’t launch a demo looking for a use case.
What Builders Should Do Now
- Start with a specific pain point, not “an assistant for everything.”
- Design permissions to expand gradually. Let users give the agent more access as it earns trust.
- Treat security as a launch requirement, not a patch to ship later.
- Before you build hardware, ask whether a phone app would solve the same problem.
Fadell’s message for this moment is that AI’s capability is way ahead of people’s willingness to trust it. Whoever closes that gap first, whether that’s Apple, Meta, OpenAI or a startup, will set the terms for personal AI. TechCrunch AI has the full account of his talk.