Tony Fadell on why the first wave of AI gadgets failed — and what comes next

The Novelty Trap

We have spent the last eighteen months watching a parade of plastic rectangles and circular pins try to convince us that the smartphone is dead. It isn't. Tony Fadell, the guy who actually helped kill the old world by building the iPod and the iPhone, recently pointed out the obvious: the first wave of AI gadgets failed because they were solutions looking for a problem.

As a founder, I see this mistake constantly. We get so enamored with a new capability—in this case, Large Language Models—that we forget to ask if anyone actually needs a standalone device to access it. Fadell’s critique hits home for anyone building in the hardware space. Most of these first-gen devices were just expensive wrappers for a ChatGPT subscription. They were slow, they hallucinated, and most importantly, they didn't do anything your phone couldn't do better.

Hardware is Unforgiving

Building hardware is a nightmare compared to software. If you ship a bug in a dapp or an AI agent, you push a patch. If you ship a piece of hardware that gets too hot, drains its battery in two hours, or fails to connect to the cloud, you have a very expensive paperweight. The first wave of AI wearables ignored the basic physics of user experience. They traded reliability for the 'magic' of voice interaction, but the magic wore off the third time the device failed to understand a basic command.

Fadell notes that these devices lacked a 'reason to be.' In the early days of the iPod, the reason was clear: a thousand songs in your pocket. The iPhone was a phone, an internet communicator, and an iPod. The current crop of AI pins and pendants are... what, exactly? A way to avoid looking at a screen? That’s a noble goal, but it’s not a product. It’s a feature.

The Trust Deficit

Beyond utility, there is the massive issue of trust. We are asking people to wear microphones and cameras that are constantly feeding data into a black box. For builders, this is the biggest hurdle. The first wave didn't just fail on technical specs; they failed to explain why we should trust them with our private conversations and surroundings.

Fadell emphasizes that the next wave needs to be 'proactive, not just reactive.' If an AI device only speaks when I speak to it, I might as well just use my phone. For hardware to earn a place on our bodies, it needs to anticipate needs without being intrusive. That requires a level of edge processing and privacy-first architecture that most current startups are simply bypassing in favor of cheap cloud API calls.

What Builders Should Focus On

If you are building in the AI or crypto-hardware space right now, stop trying to replace the phone. The phone is the hub of our digital lives and it’s not going anywhere soon. Instead, look for the gaps. Look for the moments where pulling out a phone is socially awkward or physically impossible. That is where the hardware opportunity lies.

• Focus on specialized utility: Don't build a general-purpose AI buddy. Build a tool that does one thing—like transcription, translation, or health monitoring—ten times better than an app.
• Local execution: Privacy isn't just a marketing buzzword; it’s a technical requirement. If the data doesn't have to leave the device, the user trust increases exponentially.
• Battery life is a feature: If it can't last a full day, it’s not a wearable; it’s a chore.

We are currently in the 'PDA' phase of AI hardware. Remember the PalmPilot or the Blackberry? They weren't perfect, but they paved the way by solving specific professional problems. The 'iPhone moment' for AI hardware will only happen after we clear out the junk that treats the user like a beta tester for unproven LLM wrappers.

The Founder Perspective

I’ve talked to dozens of founders who think putting a specialized chip in a necklace is a billion-dollar idea. It’s not. The value isn't in the form factor; it’s in the invisible friction it removes. Fadell’s perspective is a reminder that we need to be skeptics of our own creations. We need to ask: 'If the AI hype died tomorrow, would this device still be useful?'

For most of the first wave, the answer was a resounding no. The next wave of builders needs to be obsessed with the problem, not the model. We need devices that act as agents on our behalf, securing our data through decentralized protocols and providing actual value that justifies the extra weight in our pockets.

The first generation of any technology is usually about showing what is possible. The second generation is about showing what is necessary.

We are moving into the necessity phase. The shiny toys are breaking, and the real work is beginning. If you're building, focus on the boring stuff: latency, privacy, and utility. That’s how you win the long game.

The Takeaway

AI hardware isn't dead; it just needs to grow up. Stop building accessories for a hype cycle and start building tools for humans. Trust is earned through consistent utility, not flashy keynotes. The winners won't be the ones with the best marketing, but the ones who solve a problem so well that the technology becomes invisible.

Read the original at TechCrunch AI →