I joined Cyanogen at the strangest moment in its life: right after it shut down the phone software that made it famous, and just before it told the world it had become a self-driving company. My job was the software that ran the vehicles.
Cyanogen started as a rebellion. CyanogenMod was a community-built version of Android, made by Steve Kondik in 2009, that grew to run on tens of millions of phones by giving people control the manufacturers would not. In 2013 it became a company, raised more than 110 million dollars, and set out to do something audacious: take Android away from Google. Its CEO told Forbes in 2015 they were “putting a bullet through Google’s head.”
They lost. The deals fell apart, the money ran out, and at the end of 2016 the company shut down the phone software entirely. The open-source project forked and walked away. What was left was a hundred-plus engineers, a war chest of hard-won expertise in making software run on other people’s hardware, and no product.
That is the company I joined.
The bet was quietly brilliant: the same skill that let Cyanogen bend Android onto any phone could let it bend autonomy onto any vehicle. Take the sensors, the compute, and the software that make a car drive itself, and make them a kit you could fit to hardware someone else already built. They called it DriveMod. When I arrived in early 2017 the pivot was not yet public; that reveal came months later.
I led product for the part an operator actually touches: the console that watches a fleet of autonomous vehicles and keeps everyone around them safe. The design team had been let go in the cuts, so I ran design too, which is why the screens on this page are mine, drawn as well as specced. And because a company rebuilding a platform from scratch buys faster than it builds, I helped steer the acquisition of smaller startups whose technology we folded into DriveMod, deciding what to buy, what it would become inside the product, and how it fit the roadmap.
A self-driving fleet is only as safe as the person who can see all of it at once. The console I led is that view: a live map where every autonomous shuttle is a moving point, every route a drawn path, every stop a waypoint, and every rider a number that has to add up. Select a vehicle and you see exactly what it is doing, “EV 2453, three of six, en route Terminal 3, fourteen minutes, forty percent charge.” The rail beside the map answers the only two questions that matter in operations: where is everything, and is anything wrong. It had to stay calm and legible under real load, because someone would watch it all day while machines moved people around with no driver.
The map is the product. Everything else is there to tell you the map is fine.
You test a driverless vehicle carefully, in the real world, or not at all. Three pilots, told plainly.
You do not start with a car at speed. You start slow and low-stakes. Out of our Palo Alto headquarters we tested on street-legal electric golf carts, the same class of vehicle Cyngn took a California permit to drive autonomously that year. Slow enough to be safe, real enough to be true.
To learn how the vehicle should see, we prototyped computer vision on Teslas, running the early deep-learning models on NVIDIA Jetson modules, the small embedded AI computers that were, at that exact moment in 2017, becoming the brains of self-driving prototypes. Teaching a car to recognize a person, a curb, a path, on hardware the size of a paperback.
The most interesting problem was a port. Singapore had automated much of its shipping, and with the cargo handled by machines, the biggest remaining source of injury was people, on foot, moving around enormous autonomous equipment. We began prototyping with the Port of Singapore on autonomous people-movers for Pulau Brani: shuttles that could carry workers safely through a yard built for robots. The fleet console was built to watch exactly that.
The Pulau Brani map in the console above is this last pilot’s record: the port, the terminals, the shuttle routes.
Alongside the live console I worked on a reporting tool that took the same language, the map, the status colors, the telemetry cards, and turned it toward data at rest: jobs, coverage, alerts, and city-scale patterns laid over a map. Live operations and after-the-fact insight, one visual grammar.
I moved on later in 2017. Cyngn kept going. The company that could not win the phone found its footing in vehicles, took its self-driving story public that October, and in 2021 went public on the Nasdaq. Today it builds autonomous industrial vehicles, tuggers and forklifts that move material through factories and warehouses, for customers like John Deere. The rebellion that tried to take Android from Google became, of all things, a real autonomous-vehicle company. I was there for the turn.
Some companies get a second act by staying the same. This one got its by throwing everything out and keeping only what it was good at: making autonomy run on someone else’s machine.
