Just nine months after shipping its AI Pin wearable gadget, Humane is shutting down. 

The AI startup announced yesterday that it struck a deal with HP to acquire its assets for $166 million. The Humane team will join HP to form HP IQ, the firm’s new AI innovation lab focused on building an intelligent ecosystem across HP’s products. It’s probably the best possible outcome for Humane and its investors who poured over $240 million into the short-lived company (including Microsoft, Open AI CEO Sam Altman, Salesforce CEO Marc Benioff, and others), since the writing has long been on the wall. 

For customers who shelled out for the pricey gadget, however, it’s somewhat of a loss. In addition to taking the AI Pin off the market, the company closing shop means already-purchased devices will cease to work at the end of the month. The company is urging customers to download any photos, videos, and notes off their devices before they’re permanently deleted on Feb. 28. But the truth is—basically no one was enjoying the AI Pin anyway. 

Humane was trying something new, and it deserves credit for that. But the whole saga is a lesson about the risks of shipping products that are not ready and do not deliver on their promise, as has emerged as a trend in the fast-paced AI era. 

The downfall 

From the moment it was demonstrated in a stark announcement video, the AI Pin—pitched as a screen-less, app-less wearable device for interacting with large language models—garnered intense skepticism. Reception only worsened when reviewers got their hands on the product, with many reporting that it simply did not work. Notable tech YouTuber Marques Brownlee famously called it “the worst product I’ve ever reviewed.”

The public didn’t receive the product any better. In the months after launch, daily returns of AI Pins were outpacing sales, The Verge reported. In October, five months after the product shipped, the company slashed its price in an attempt to drive sales, charging $499 (for a version that didn’t include the original accessories) down from $699. Of course, users would still have to pay the $24 subscription fee every month. Not only did the Pin not work well, but it wasn’t cheap.

The impacts of the AI ‘race’

While there will always be a learning curve with first generation products and risks associated with paying a high price to be an early adopter, there is some level of expectation that a product being sold will generally do what it’s supposed to do. As already discussed, the AI Pin did not even come close. 

“The AI Pin is an interesting idea that is so thoroughly unfinished and so totally broken in so many unacceptable ways that I can’t think of anyone to whom I’d recommend spending the $699 for the device and the $24 monthly subscription,” wrote The Verge’s David Pierce after spending two weeks testing the device. 

Users reported that the AI Pin often failed to complete tasks and answer queries, or took forever to complete them when it did. They also reported issues with the accuracy of the information provided by the AI Pin, its interface, the resolution of the projector, battery life, and pretty much every aspect of the device. At one point, the company even warned owners not to use the charging case after discovering it may pose a fire safety risk. 

It’s fair to wonder: Why didn’t Humane continue to work on the product until it performed much better? 

While the company was founded in 2018 and has been working on the product for years, the launch of ChatGPT in November 2022 turned the previously experimental and research-driven field of AI into a fast-paced commercial “race.” Now, progress in model capabilities isn’t happening over the course of years or even months, but rather weeks, and launches are happening nearly every day. Humane was taking its time for a while, but time ran out.   

In addition to the pressure to launch quickly and beat the competition in the post-ChatGPT AI era, the costs associated with AI R&D only exacerbate the issue. Developing and running AI products is very expensive, putting startups in particular in tight situations and timelines. 

During a recent conversation about how Rabbit, the company making a similar LLM-in-a-box device, also launched an unfinished product and has been essentially building the device in public, CEO Jesse Lyu told me the company simply doesn’t have the the runway or resources of a tech giant that would enable it to take more time. “We have to make sure that we take our shot and move fast. This is the only way that we can stay in competition,” he said.

There’s also just the fact that the underlying AI technology is still very much being developed, and all AI models (Humane used an ensemble of AI including ChatGPT and Gemini) are plagued by mistakes and hallucinations. Humane was the first consumer hardware launch of the AI era, and well, maybe there is a reason there hasn’t been too much interest in spinning LLMs into gadgets just yet. 

And with that, here’s more AI news. 

Sage Lazzaro
sage.lazzaro@consultant.fortune.com
sagelazzaro.com

AI IN THE NEWS

NIST prepares to cut AI Safety Institute and CHIPS employees. As the Trump administration continues its sharp cutbacks of personnel working at U.S. federal agencies, the National Institute of Standards and Technology is gearing up to make cuts that could gut the AI Safety Institute and CHIPS for America program. NIST is preparing to cut 497 people, Axios reported. It’s the latest move from the Trump administration signaling its disinterest in requiring any sort of safety testing or regulations for the creators of AI technologies.  

Google launches an AI Co-Scientist based on Gemini 2.0. Described as a collaborator for scientific researchers, the multi-agent AI system is designed to generate novel hypotheses, write research proposals, help scientists review literature, and accelerate the speed of scientific discoveries. It relies heavily on reasoning abilities as well as self-critique, using an included feedback tool to refine its own hypotheses and proposals. Research organizations can now apply to gain access to AI Co-Scientist. Read more from Reuters. 

Microsoft unveils an AI model that can generate gaming environments. Called Muse AI and detailed in a paper published in Nature, the model was trained on human gameplay data from the Xbox game Bleeding Edge. The company says it understands the physics of a 3D gaming world and can react to how players interact with a game. For now, Microsoft is exploring how the model can be used to reconfigure classic games for modern hardware, but the idea is that it could eventually be used to help game developers create parts of games.

FORTUNE ON AI

Amazon’s big bet on warehouse robots is already getting a boost from generative AI —by John Kell

AI safety advocates slam Trump administration’s reported targeting of standards agency —by David Meyer

The most underreported and important story in AI right now is that pure scaling has failed to produce AGI —by Gary Marcus

MLB will test robo-umpires in spring training games starting this week —by Chris Morris

How OpenAI will fend off Elon Musk —by Andrew Nusca

AI CALENDAR

March 3-6: MWC, Barcelona

March 7-15: SXSW, Austin

March 10-13: Human [X] conference, Las Vegas

March 17-20: Nvidia GTC, San Jose

April 9-11: Google Cloud Next, Las Vegas

May 6-7: Fortune Brainstorm AI London. Apply to attend here

EYE ON AI NUMBERS

400

That’s about how many predictions made with Crunchbase’s new AI-powered prediction engine came true last month, according to the company. 

The company, a longtime database for information about private companies, tells Eye on AI the solution analyzes billions of market signals to predict trends and major events across millions of private companies with 95% accuracy. Among its predictions that came true last month were Innoscience going public (predicted two months in advance), Grammarly’s acquisition of Coda (predicted two months in advance), and Hippocratic AI’s funding round (predicted 4 months in advance), according to the company. Crunchbase is reconfiguring itself around the new capabilities, claiming that “historical data is dead” and now pitching itself as a AI-powered prediction engine for startup funding rounds and other events.