Tony Fadell says early AI gadgets missed the point

Speaking at MIT Future Fest, the iPod co-creator said first-wave AI devices chased novelty, not real needs, and failed to earn user trust.

Tony Fadell used his appearance at the first MIT Future Fest to deliver a blunt verdict on the first wave of AI gadgets. He pointed to the Rabbit R1, Humane Ai pin and Limitless pendant as devices that attracted attention more as interesting tech than as products that fit naturally into everyday life. According to IT-PUB News, his criticism centered on two connected problems: weak real-world use cases and a lack of trust. That matters well beyond a few gadgets, because these were among the most visible attempts to turn generative AI into consumer hardware.

His comments also land on a broader industry fault line. An AI assistant may sound compelling in theory, but people are unlikely to rely on it unless it solves a clear problem and handles sensitive data safely.

Fadell says the first AI devices lacked a real use case

Fadell said the companies behind some of these early devices had contacted him for advice, but he chose not to get involved. In his view, they were all making the same basic mistake: the products did not address a meaningful pain point.

He described the Rabbit R1, Humane Ai pin and Limitless pendant as “Gen 1” AI products aimed more at tech enthusiasts than ordinary consumers. They looked novel, he said, but did not meaningfully connect to daily life.

The criticism was not just about design or marketing. Fadell’s point was simpler and harsher: a product can generate plenty of buzz and still fail if people do not immediately understand why they need it.

Trust may be the bigger challenge for AI assistants

Fadell also focused on a second obstacle — trust. Early AI devices promised something close to a personal assistant, but that is a hard pitch for people who have never had a human assistant in the first place.

He said only a tiny share of the world’s population has ever worked with a human assistant. That means most consumers do not have a clear model for how an assistant should behave, or how much they should rely on one. In his telling, the jump to an AI helper is much bigger than many companies assume.

To make the point, he compared it with hiring a human assistant. You would not hand a new assistant access to your bank account on the first day, he said. That kind of confidence takes time to build, and he argued the same is true for AI systems that may eventually manage calendars, messages or financial tasks.

That is a serious obstacle. The most useful AI assistants would need access to highly sensitive information, and without trust, the product stays a novelty instead of becoming a daily tool.

Security concerns are already shaping the debate

Fadell’s remarks arrived at a time when AI assistants are already facing security scrutiny. He pointed to Meta’s Muse, launched as an all-purpose AI assistant, as an example of why users may hesitate.

According to the source text, a security researcher quickly found a serious vulnerability in Muse after launch. It also cites a recent 404 Media report saying some Meta employees found security issues before launch, which led multiple teams to rush fixes and work overtime.

Those details help explain why trust is becoming central to both AI hardware and software. If a product is supposed to handle personal information, even early security problems can damage confidence before the service has a chance to become part of everyday routines.

Fadell summed that up by saying trust and safety will be “paramount” for anything people hand over to an intelligence system. He added that the only company he could currently imagine doing this well is Apple, or possibly one other company.

Fadell argues successful AI will need to run on-device

Another major theme in Fadell’s comments was where AI should run. He predicted that a successful AI agent will need to operate on-device only, rather than relying heavily on the cloud.

He tied that view to both privacy and practicality. Keeping data on the device, he suggested, reduces the need to send sensitive information elsewhere and can make the technology lighter. He also pushed back on the idea that huge data centers will be the main answer for AI assistants, saying he does not believe that path will dominate.

Fadell argued that modern phones and devices already have a lot of computing power. In his view, the real question is not whether devices can do more locally, but whether companies will design products that use that power without demanding too much personal data.

Apple, he said, has built trust with consumers by keeping biometric data on-device in features such as Face ID. At the same time, he noted that Apple does not have its own top-tier AI model, and that the new Siri AI runs on custom-built versions of Google’s Gemini.

Hardware access may be driving the new AI gadget push

Fadell suggested that companies such as Meta and OpenAI are moving toward dedicated gadgets partly because they do not have the same hardware footprint as Apple. Apple, he noted, already has billions of devices in circulation.

His argument was that companies without that kind of hardware reach may try to build their own devices in order to collect the sensor data they need. He described a process in which a device asks for access to video, audio and location, among other things, and gradually ends up collecting a long list of permissions.

In that scenario, he said, companies may create screenless devices that connect to a phone through Bluetooth or Wi-Fi and then reach the web or a 5G network. As he framed it, the goal is to gather the sensor data they cannot easily get through a phone alone.

That helps explain why so many AI hardware experiments have appeared in a short period. It is not just about putting AI into a gadget. It is also about finding another way to access user context and data.

Startups have much less room for a miss

The event also touched on the business reality behind AI products. Asked about luck and product-market fit, Fadell said startups face a much harder path than larger companies because they usually get only one real chance.

A failed product can be fatal for a startup, he said, unlike at Apple, where even a product such as the Vision Pro can be treated as one step inside a much larger portfolio.

That may be one reason his comments drew so much attention. Fadell was not simply dismissing a few AI gadgets — he was describing the conditions that may determine which AI products survive. In his view, novelty is not enough. The products that last will need a clear purpose, strong security and a level of trust that many AI devices still have not earned.