The other night I checked my phone and I had a Facebook Messenger notification from Emil Mikhailov, cofounder and chief executive of 1-year-old startup Xix.ai. Attached was a screenshot of his Android homescreen. In the middle was a card showing his meeting the next morning with a prominent Silicon Valley investor. Beneath it were buttons for directions to the meeting and the email thread associated with the calendar event. And still below that were eight apps that the launcher thought would be most relevant to him at that moment.
“I absolutely love it!” wrote Mikhailov. “It shows me my upcoming event and associated action items.”
By “it” he was referring to Xix One, the Android launcher he and his team developed. The screenshot showed an early version of Xix (pronounced X-I-X — or you can just call it 19) running on his Samsung Galaxy S7. He hated Android; he missed talking with his wife, Anna, on iMessage. Sure, he told me, there was absolutely room for the launcher to get smarter — he thought it would have been better for Xix One to show him the card right before he left for the meeting with the investor, rather than the night before. But still, it was a lot more relevant than the default grid of apps that you normally see. Plus, the launcher was already smarter than it had been in the previous days.
If Mikhailov and team have their way, all of us with Android phones will have this type of very personalized content in our hands, at the ready. Very soon.
Next week Xix will present at startup accelerator Y Combinator’s demo day. It will join more than 100 startups pitching investors. Mikhailov hopes to find a match with investors excited by the team’s long-term plan to radically improve mobile-first personalization.
After all, they have a real product that works. But they want to go further. They basically want to provide a smart assistant to help people get things done more efficiently. Not like Alexa, Cortana, Google, Siri. Something different. Something better.
Something that will draw on readily available data sources and contemporary artificial intelligence (AI) systems. But the system Mikhailov foresees will work in a way that today only really exists as an idea. Maybe it will work, and it will become pervasive because of its value. Maybe it will just be a bright idea in the mobile ecosystem that becomes just a footnote to a Wikipedia article about Android launchers. But heck, it’s interesting.
From chance encounter to a beloved corgi
A year ago none of this existed.
Mikhailov, the founder of one of the Bay Area’s AI meetup groups, was just another guy hanging out at a coffeeshop — specifically Epicenter Cafe in San Francisco’s tech-heavy SoMa neighborhood. While nursing his double espresso, he looked at the person in front of him and saw a guy hunched over a laptop, reading an academic paper, in a T-shirt that said something in Russian about a St. Petersburg programming competition. The paper was “Maximum Entropy Deep Inverse Reinforcement Learning,” by Markus Wulfmeier, Peter Ondruska, and Ingmar Posner.
“Hey, I read this paper recently,” Mikhailov told the guy, whose name is Alex Skidanov. Skidanov ended up helping him run the meetup group. Within months they were talking about startup ideas.
It’s almost a scene out straight out of central casting: a pair of Russian-speaking dudes who are into AI. Mikhailov, 27, is more of the operations and business guy, and Skidanov is the guy who writes the code. Mikhailov speaks thoughtfully, slowly; Skidanov, 30, shoots out one sarcastic line after another. Together they complement each other as a founding duo.
A few months ago Skidanov felt the time had come to move on from MemSQL, where he had spent five years, most recently as director of engineering. He spent a couple of months as a consultant to nonprofit AI research startup OpenAI. Then he officially joined Xix. They decided they would build an assistant for Android. So by necessity, Skidanov went to work familiarizing himself with Android, sitting alongside Mikhailov in a SoMa coworking space, the same place where they always host their meetup events.
The first thing they built was a recurrent neural network (RNN). They knew what they wanted to create, but they didn’t have a name for it. So they scraped AngelList and took the names and descriptions for startups and used it to train the RNN, and then they had it produce some names. One of the first ones was Xix. They thought that looked cool. But they wondered what they would say when people would inevitably ask them what it means. As a joke, they figured they could say it was 11 times — or XIx — better than the closest competitor.
Their second effort was a nerdy-looking website that looked like a terminal window: black background, lines of text in green. At least it did something — it conveyed that they were about building cool technology. It even gave a clue of what the startup was up to if you rolled up your sleeves and typed
cat about.txt into the command line. “Using latest advances in deep learning to build a next-generation mobile AI OS for augmenting human intelligence,” it spat back.
In November they got into Y Combinator. To make sure they didn’t get distracted, they moved into the SoMa apartment where Mikhailov lived with his wife. They spent a lot of time figuring out what they needed to do to build what they wanted to build.
In January, Xix released an app called Jyr Assistant in the Google Play Store. The app decorates your homescreen with a cute little corgi, sort of like Batman, Skidanov’s real-life corgi. When you tap on the dog, which persists across apps like a Facebook Messenger chat bubble, you see a list of deep links into apps like Google Maps, Uber, and Lyft, as well as buttons that take you to an overlay showing a map of where you are, and an overlay for your calendar for the next few days. The overlays used the same theme as the Xix website: sleek green on black background.
As Jyr gained users, Xix gained information of what people were doing on their phones. Over the course of a few updates, Xix gave Jyr more capabilities.
In early February the duo released a second Android app, MotivAction. It lets you track how long you spend in certain apps, like Facebook and Twitter, and it can even limit your use of those apps if you’d like. It was a standalone app that you could jump into and out of, instead of something that was with you on your phone all the time.
And later in February, Mikhailov and Skidanov released Xix One. It drew a little bit from Jyr, because it surfaced calendar information and because it kept up the black and green theme. And it drew some from MotivAction, because its data smarts derive from what’s onscreen. But Xix One was a full-on launcher that took the place of a phone’s homescreen. There was a window with upcoming calendar events in a tiny typeface, along with a window showing your most recent emails. But there were also some smaller windows with app icons. Xix One was predicting which apps people would want to use next — and with each day it got better at doing that. At times, instead of showing app icons, it would provide deep links into content within apps — individual tweets, SMS conversations, even websites. On the bottom right was a big X icon — it stood for Xix, of course — that would let you quickly get to the phone app, the default browser, and the app drawer.
Next Polosukhin, then MAYA
Later in February, the guys got a third cofounder, another Russian speaker who was into AI: Illia Polosukhin, fresh out of Google Research. The 27-year-old had met Skidanov years earlier when the two were involved in programming competitions. Polosukhin’s first order of business was incorporating a three-layer artificial neural network into the app in order to make better app predictions. It was based on TensorFlow, Google’s own open-source framework for deep learning, a trendy type of AI.
But that wasn’t all that he did. He also whipped up a system to ask users how their day was every night before bedtime. Users could choose a smiling face, a blank expression, or a frowning face. That information sounds simple, but it can provide a feedback loop to help Xix understand what people like to happen during the day, which in turn can lead to better predictions and suggestions.
Within a couple of weeks, while visiting Y Combinator’s headquarters in Mountain View, Mikhailov got some important feedback. One partner said the app didn’t look very good, was too restrictive, and had no exit button. Another partner told Mikhailov about the MAYA Principle. MAYA stands for “most advanced, yet acceptable.” It says that you should build things that will direct people toward the future but are at the same time understandable given where those people are now. It’s why the Apple Newton never took off but the iPod and iPhone did, even as it evolved to a touchscreen that resembled a Newton. In that conversation, it became clear to Mikhailov that the Xix One launcher would be too foreign to some people and might never take off.
He went back to the startup’s office and informed Polosukhin and Skidanov that they would have to come up with a new design, something that would work well with people’s familiar Android environment, something that wouldn’t look like a terminal window. They urgently rebuilt the app and then released the new version as an update in two weeks. The new version had only slightly different functionality but offered a far more approachable user interface — as well as Polosukhin’s nightly feedback check.
Instead of taking up the entire display, the second version of the launcher uses much of the middle but leaves room for the standard Android app tray and a few additional apps on the bottom. The black and green theme was replaced by a stack of, typically, four cards you can swipe through — calendar events, the latest emails, recent contacts, and places to go with Uber or Lyft — and below that, a collection of eight “predicted apps” that Xix thinks you’ll want to use next. User feedback was clear: The launcher looked more familiar and user-friendly.
Not that the team was perfectly happy with the product. One day while I was visiting Skidanov called for everyone’s attention. Typically he kept to himself, wearing headphones, staring at a big monitor, typing coldly into a Kinesis Advantage mechanical keyboard that was hooked up to a giant PC tower running Ubuntu — but now he wanted to show us something. He grabbed his old Galaxy Note 4 phone with the original launcher on it. By pushing a few buttons on his computer, all the app icons took on a glowing green hue. He laughed a maniacal laugh. Then with the push of another few buttons, the entire desktop wallpaper in the background turned green. He laughed again, this time even more devilish. He looked up at me, grinning, and I grinned back. I liked the look myself, and I really liked how so much information was squished into the homescreen. You just had to be careful you didn’t tap on the wrong thing.
Within a week the trio was testing out a redesign that included a bright, Googley background color for the stack of cards, along with a label for each card. And that’s when Mikhailov found himself genuinely impressed with the launcher and gushed to me over Messenger.
The pursuit of magic
“One consistent thing people were telling us — they said one word, and it was magic,” Mikhailov told me one night over dinner next to one of Google’s bayside San Francisco offices. “They said, ‘I want to see the magic.’ There’s no magic in there. We realized people wanted fucking magic. What would be the magic? The magic is you have very precise, time-sensitive predictions and actions associated with them, something that you don’t need extra effort for them. You just confirm this, just confirm it for me.”
That gets at the future where Xix is headed. The three guys are building software that will be able to do more complicated tasks for you without you even asking for them. It will know what you want because it has seen you do those things before. You’ll just need to say yes, that’s what I want, and everything will be taken care of.
It will do these things by essentially farming out mobile tasks to virtual versions of your phone. It will be able to enter your long password for you. It will be able to take information from one app and enter it into another app. It will give you less stupid little stuff to do.
In other words, the driving idea behind Xix is to waste less time.
Skidanov walked me through an example of how things work today. Let’s say you want to find a good restaurant and go there. You might open the Yelp app, browse or search, choose a restaurant that looks good, copy the address to the clipboard, then open Uber, wait a few seconds until the app shows your location, paste in the address of the restaurant, adjust the text so it registers in Uber, receive a prompt to log in again since it’s been awhile since you last used Uber, so you open up Lyft instead, wait a few seconds until it’s ready, hit OK on a pop-up message, tap the “pickup” button, paste in the address, edit it, and then order the car.
“That’s not the most uncommon use case. That’s something people do,” he said.
In some cases, apps intentionally get in the way of taking direct actions in order to generate revenue — by showing you ads, for example. Xix has already found ways to circumvent that, with deep links to emails, rides to particular destinations, and so on. But there’s more to do. In the future, Xix will delegate processes to apps running in the cloud.
“In my head it all makes sense,” Skidanov said, momentarily putting aside the biting wit and conveying some of his real visionary thought.
There are a number of ways this could play out. It could be that Xix will try to bundle the software on new phones. Maybe over time the launcher will give way to a forked version of Android, Mikhailov said.
In any case, he’s come to accept that while they do have a working product, they’re building something that might not have a big payoff right away. It could take time.
“We’ll just do something that is way better than the iPhone,” he said. “I mean, it would be hard, but why not?”