The AI Gadget Died. The Lesson Didn't Land.
Two years ago the standalone AI device was going to retire the smartphone. It's worth being precise about how that turned out, because the reasoning behind it is still circulating.
Cause of death
Humane raised $230 million, shipped fewer than 10,000 AI Pins at $499 apiece, and sold its assets to HP for $116 million. HP killed the product on arrival.
Then came the detail that came to define the whole category. On February 28, 2025, the Pins stopped working. Not degraded. Not unsupported. Once they couldn't reach Humane's servers they lost the ability to make calls, send messages, answer questions or touch the cloud. People who'd paid $499 were left holding an object.
Rabbit did better, though that's a relative claim. The R1 sold around 100,000 units, then absorbed mass returns. As of this month the company is still operating, still shipping, still pushing firmware and feature updates, and reportedly struggling to make payroll.
In fairness: the R1 running today is materially different software from the one MKBHD called barely functional at launch. Rabbit kept working. The trouble is that the reviews were written in week one, and week one is when the category's reputation got set in concrete.
The pattern underneath
Line the post-mortems up and the same three failures keep showing.
The devices launched against demo-quality expectations they couldn't hit in a hallway with weak Wi-Fi. They shipped with latency and reliability problems phones had sorted out a decade earlier. And what they did, phones already did well enough.
That last one kills on its own. A new hardware category has to beat the thing already in your pocket, and beat it by enough to justify a second device, a second charger and a second subscription. Being differently good doesn't clear that bar.
The wider bill
Hardware wasn't the only casualty. By one industry accounting, three high-profile AI products shed more than $5 billion in combined value inside twelve months.
OpenAI's Sora is in that count. The same analysis estimates it burned roughly $15 million a day in compute against $2.1 million in total lifetime revenue before shutting down on April 26, 2026. Those numbers are third-party estimates, not disclosed financials, but the shape isn't really contested: real technical achievement, no path to paying for itself.
The shared diagnosis fits in one sentence. They mistook technological novelty for product-market fit.
Why the autopsy still matters
The AI gadget isn't a closed chapter. The pitch just migrated to glasses, pendants, pins and assistants that want deep access to your accounts, and it shows up with the same structure: a great demo, a monthly fee, and a device whose usefulness depends entirely on its vendor staying solvent.
Humane established what the real risk is here, and it isn't that the product disappoints. It's that the product ceases to exist. Software that dies at least leaves your data behind. Hardware that dies leaves a paperweight.
The question to put to any AI device in 2026 isn't whether the demo impresses. It's what the thing does on the day the company doesn't.
Image: Shotkit, via Pexels
