SelfieBot Was a Joke. The Prophecy Wasn't.
A small story from a small office in Boulder, and what a 2014 April Fools prank predicted about the consumer AI race in 2026.
Let me first tell you about Sphero.
March 31, 2014. Boulder, Colorado. The company that would later become Sphero was still called Orbotix, and at that moment we were essentially a robot-ball selling entity. At best, we were a specialty toy shop's way of pulling a crowd on a Saturday-afternoon street mall. Our business was heavily concentrated around Black Friday. Our mission — as a public, beloved one-trick pony — was to even out our sales cycles across the rest of the year through unnatural tactics that brought awareness and delight to our potential buyers.
April Fools' Day was one of the more natural unnatural sales cycles we created.
In 2012, we faked a bit of magic by wrapping our robot ball in an orange peel and "driving" it around the grocery store. In 2013, we feigned a fake Kickstarter to back our Peacekeeper — a 3-foot, 150-pound rolling sphere that crushed everything in its path. We listed it for $10,000. People actually tried to buy it. We had to spin up a real Indiegogo campaign to sustain the bit.
In 2014, it was SelfieBot.
A fully autonomous flying drone-ball that locks onto your face and follows you everywhere — to bed, to brunch, to the bathroom mirror — patiently waiting for the moment you tilt your chin to a "flattering 35-degree angle" so it can capture the perfect selfie. The tagline: "Always watching for life's most precious moments." It felt like something out of a Black Mirror episode before the Netflix series even existed.
We put up a reservation page at gosphero.com/selfiebot. The video went live the night of March 31. Within hours, real outlets were covering it as if it might be real. Brad Feld — Foundry Group, board member, fully in on the bit — blogged about it as a genuine product launch: "Selfiebot is an autonomous flying robot that shoots HD photos of you, freeing you from the limitations of a handheld smartphone when taking selfies. Check out Selfiebot today." Beautiful. Total commitment.
The internet laughed. The team moved on. Today the video has over 323,000 views. It's still up on YouTube, unlisted.
It was a perfect prank because the joke wasn't that the product was absurd. The joke was that the product was plausible —
The selfie was the cultural protagonist of 2013. Oxford named it Word of the Year in November. Usage of the word had increased 17,000% in twelve months. The cultural panic about narcissism, surveillance, and manufactured self-image was at full volume. We took the two technologies that defined that exact moment — drones and selfies — smashed them together, and built a product that was funny because it was too honest.
What people really enjoyed about our pranks is that they weren't too far off from reality. Peacekeeper was a thesis on scale — what happens when you make a hit consumer toy bigger until it becomes dangerous. SelfieBot was a thesis on attention and surveillance — what happens when the technology decides it knows what's worth capturing better than you do.
But SelfieBot hit different.
Because the four technologies it satirized — augmented reality, drones, imaging, and robotics — were the four technologies I had been working in, in some configuration, for my entire career:
Parrot, 2011. I launched the first iOS app-enabled consumer-grade drone at CES. The hardware was elegant. The marketing was tight. The thing it could not do — the thing none of the drones in market could do at that point — was know who you were.
Sphero, 2014. We made up a flying robot that knew exactly who you were and locked onto your face. The hardware was a joke. The behavior was the thesis.
Snapchat, 2015. I joined the hardware team to lead GTM. The team was deep in research on the history of cameras, imaging, and mixed realities in the real world. I went on to launch their first camera product — Spectacles. The camera came down out of the sky and landed on your face.
During my time at Snapchat, I showed Evan our SelfieBot campaign. We laughed. A few years later — after my tenure — Snap uneventfully released Pixy: a pocket-sized, free-flying camera that lifted off your palm, captured you from the air, and landed back in your hand. I won't claim SelfieBot was the seed for Pixy. But it wasn't not the seed either.
Three jobs. Three form factors. The same problem getting closer each time: how does a piece of hardware see, remember, and serve up the moments of your life without you having to point at them?
—
Then the camera tried again. This time as glasses.
Meta has been working on smart glasses since the moment Spectacles launched. Ten years later, I'm playing beach volleyball in Hermosa wearing my Ray-Ban Metas — and honestly, this is the most compelling niche use case for face cameras to date. There is no other moment in my day where I want hands-free, head-pointed video capture more than when I'm in the middle of a rally. And the replays are great for coaching - I can see where my hands are on the ball [and aren’t].
That said. My honest review of the Ray-Ban Metas is marred by three annoying user experiences that have me wearing them as sunglasses and not as a face camera:
- The Bluetooth handshake is greedy. If I have my AirPod Pros on with noise cancelling, the Metas periodically re-pair and take over. I have to ‘Forget’ the device and re-pair when I want to use them.
- Within an hour of wearing them on the beach, both the glasses and my phone are dead. The energy budget is brutal.
- The captured content downloads to a separate app, and reviewing the footage is cumbersome enough that I rarely do it. Which means it might as well not exist.
So we're still not there. The drone version, the glasses version, the pin version — none of them have closed the loop. Which brings me to where I've actually been spending my time.
—
Most recently, I've been working on a multi-modal agentic model that bridges the gap between the physical and digital worlds. As part of that work, I've been spending time with founders building in adjacent territory. The most interesting one I've come across is Memories.ai. I've spent a month talking with the founder this year about their research and technology.
They’ve taught AI to watch videos the way a human does — video as the primary input modality, not a postscript to text. Memories.ai is built around what they call a Large Visual Memory Model (LVMM): a system designed to process enormous volumes of video footage and build a searchable, contextual archive of someone's actual life. And they've paired the model with a hardware form factor — an AI pin called LUCI that continuously captures 4K video, processes it in the cloud, and makes the result queryable. And the 8-hour battery life is beyond impressive.
This is not the first AI pin. The Humane pin raised an initial $160 million and attracted what I'd now call a healthy skepticism around new form factors. Having been at the helm of bleeding-edge consumer hardware multiple times, I know the math: you get a lot of criticism before you land on product-market fit. It's the price of going first.
There are also new pin-like rumors emerging from the OpenAI / Jony Ive partnership. (Altman was an early investor in Humane.) The rumors say they're ditching the pin-on form factor for a practical instrument / pen — one that listens and writes.
—
LUCI states that it’s the ‘third core device’. I like the ambition of this statement, but I wonder where my personal robot will rank in the near future.
Right now research estimates that there are over 1,000 robot companies in the world. The vast majority of the capital flowing into the category is going to vision, dexterity, and mobility — essentially the software utility on top of moving hardware. A new generation of physical robots is about to land in our homes, and they will arrive with extraordinary general capability and zero specific understanding of you.
So the question I find myself returning to is the middleware question:
How do we get a robot to do a task the way you would do the task?
Is it like downloading your iPhone apps directly onto the robot? Is it like handing the robot your phone? Is it like setting up an entirely separate phone plan, but for your robot? Which smart-home devices become irrelevant the moment a household robot has vision? (My guess: most of them. A smart refrigerator is just a robot's first glance.)
This is the gap I'm obsessing over — an AI for doing the personal, real-world tasks that make life feel annoying, confusing, or stressful. And it's also the gap I see lifelogging hardware finally able to close. Because the most interesting thing about a continuous-capture camera isn't that it remembers your environment. It's that it remembers how you operate within it.
The robots aren't going to be trained on our environments. They're going to be trained on us. Not on what we want. On how we do it.
The SelfieBot prank predicted a clever form factor for surveillance and the ability to learn autonomously from passive daily life activities.
We laughed in 2014 because the prank was too on the nose. I think it still is.