Why New Tech Products Fail
The Humane AI Pin offers some compelling lessons.
Like many other people, I was excited by the announcement of Humane’s AI Pin a few months ago, and immediately dropped $780 for a preorder. Although I was skeptical that the device would be great on the first iteration, I believe in living in the future with the latest tech, as a way of helping to see what the next horizon might look like.
Perhaps not surprisingly, however, the initial reviews of the Humane AI Pin have been consistently negative. The device is too slow, it fails at many basic tasks, it hallucinates, and it lacks important features. To be fair, I don’t think all of these problems are Humane’s fault — the technology just isn’t there yet, but given another few years, perhaps it will be. (Some of Humane’s problems are self-inflicted, though — for example, I see absolutely no reason for the voice used by the device should be so robotic, given the amazing quality of modern TTS models.)
In thinking about what makes a new technology product successful, it’s worth looking at history. Take, for example, the original Apple II, or the Blackberry, or the iPod. All of these devices relied on cutting-edge technology (for their times), yet, they managed to become wildly successful despite all of their obvious limitations. And yet other products, like Google Glass, Google’s Stadia cloud gaming console, and the Lytro light-field camera all failed. Why?
A simple test: The 10x use case that matches the tech’s capabilities
The wrong lesson to take away from Humane’s AI Pin is that the tech isn’t ready. I don’t think this is it. There are plenty of successful products that were successful despite tech limitations. The trick is that (a) the use case has to be really compelling despite those limitations, and (b) the tech has to work well enough for the use case. Humane’s pin fails both tests. So did so many other failed bleeding-edge tech products. This seems like such a simple rule that I’m surprised more tech companies (ahem, Google) don’t keep it in mind when conceiving new products.
Let’s take early PCs, like the Apple II, as an example. Even given its terrible specs, not-cheap price tag, and severe limitations, the Apple II did one thing so amazingly well that it became a game changer for countless people: VisiCalc. Compared to using paper-based records and performing calculations by hand, VisiCalc is a 10x (or maybe 100x or 1000x) improvement. Likewise, the original Macintosh made true word processing and desktop publishing possible. The alternative, at the time, was a typewriter. It’s not hard to see that, despite how primitive they were, early PC-based word processors were still 100x better than a typewriter.
Likewise, the Blackberry was wildly successful despite its problems. Without a Blackberry, at the time, the only way to get access to your email would have been to sit in front of a PC, more likely than not on a dial-up connection. As crappy as the original Blackberry looks today, it was still super compelling and a good fit for what the current tech could achieve.
The original iPhone, notably, had no apps apart from those built into the OS, yet the overall experience (email, web browsing, phone calls, music all on one device) was super compelling despite that severe limitation.
On the flip side, where did, say, Google Glass go wrong? Like the Humane pin, it lacked any kind of 10x use case. Being able to get notifications or take pictures from your face was just not compelling enough, especially given the terrible form factor of the device. (Wearable smart glasses that look like, well, glasses, have fared much better in the market.)
In the case of Stadia, the problem was that the only use case was identical to a PC or an Xbox or a PS3 — playing games. The price of gaming on a Stadia being marginally better than an Xbox does not make up for the latency problems and lack of a decent game library.
The Lytro camera offered the ability to adjust the focal point of an image after it was taken, which is, at best, an extremely niche use case — certainly not 10x, not even for serious photographers. Further, whatever benefit one might have derived from the light field sensor was offset by the poor overall quality of the image.
It’s not just about killer apps
A lot has been written about “killer apps”, but I think there is often something lost in the definition of a “killer app”. Too often killer apps are described as merely a complement to the underlying hardware or tech that they run on. Another thing that the killer app model tends to ignore is that the app has to be 10x better than the status quo, whatever the status quo is. It’s not enough for a killer app to be something that you can get somewhere else, maybe for cheaper, or somewhat lower quality — the killer app has to be so compelling that it has to be the only way to do that job.
No matter what the killer app or the device is, there has to be a strong overlap between the key features of the app and the capabilities of the current state of technology. VisiCalc did not need better specs or fancy color graphics or anything that the Apple II could not deliver. Playing music on the original iPod didn’t need anything better than a basic button-based interface and LCD screen. Had those applications required much fancier hardware specs or a multi-touch display to be usable, there’s no way they would have been successful.
Where Humane went wrong
The Humane team seems to have forgotten these fundamentals in the design of their AI Pin. The bar to compete with a smartphone — which everyone already carries, and is much higher quality than anything the AI Pin delivers — is very high. There’s no use case for the Humane pin that can’t be done with a smartphone, even if using a phone is a little more clumsy at times (and, arguably, phones are simply better at most of the things the AI Pin is trying to do).
It’s also clear that the Humane team is trying to do things that are simply not possible with today’s tech. Hallucination-free, low-latency LLM-powered question-answering, powered by voice, just isn’t there yet. Wearable, always-on machine vision (for the AI Pin’s gesture detection) isn’t there yet.
What would a successful AI Pin product look like in 2024? I could imagine Humane developing a wearable smartphone accessory that provides the basic camera, voice control, and projector capabilities of the existing AI Pin, but mates with a smartphone that provides most of the, well, smarts. Such a device could last all day on a single battery charge and, based on the BOM cost of the Apple Watch, should be something you could sell for around $300 — and no need for a separate $24/month data plan as the current AI Pin requires.
Such a product would undoubtedly be more successful than the ill-fated AI Pin, but it still doesn’t have a 10x use case. One possible 10x use case for a pin-like form factor device would be a personal memory assistant. A wearable device that could watch and listen to everything I do and help me recall that information would be a game changer. I’m not sure that the tech is quite there yet, however.
Humane’s problem here seems to be one of hubris — thinking they could completely replace smartphones with the first version of their device. That’s nuts, and one wonders why Humane went that route.
