2024 – now
Scaling intelligence, carefully.
I build the systems underneath AI. Nothing I've worked on has moved this fast or counted for this much. The job is to make it scale without breaking the things that matter.
Shu
The things we build are starting to outgrow us. I work on the part that has to stay , , and while they do.
2022 – 2024
The library behind the screen.
At Apple I led engineering for the services behind its media: the catalog of every book, song, and show, and the systems that deliver the right one to a billion devices. Nobody notices when it works, which is the point.
2021 – 2022
Starting from zero.
I left a big company to start a small one, out of the Allen Institute for AI. I talked to customers, built the prototypes, pitched investors, and led engineering as we laid the foundations. The company was later acquired. I learned how much of building is deciding what not to build, and I learned when to walk away.
2018 – 2021
New ways to capture.
At Facebook I led teams across photo, video, augmented reality and hardware, inventing new ways to capture the world: pictures you could look around, new video formats, effects laid over what the camera saw, and the devices that shot them. 3D Photos alone reached hundreds of millions of people. It was the first time I had invented formats.
2012 – 2018
The shape of a person.
For years at Facebook I built the systems that hold who you are online: your timeline, your identity, the graph that connects you to everyone else. For a while I was the only engineer on a system that carried eight percent of the time people spent on Facebook. That is no way to run anything, so I learned to build a team around it.
2010 – 2012
Finding the signal.
I left a big company for two small ones. At Redfin I built tools that turned transaction data into decisions. One of them made agents' track records public, which the industry did not enjoy. At Wavii I helped build a service that read the news and turned it into feeds, until Google bought it.
2002 – 2007
Silicon, then neurons.
My first work was close to the metal: firmware for video chips, simulation models that cut days off a tape-out, trips to Japan to make a display controller ship. Then, at the Center for Theoretical Neuroscience, I built the interface for Nengo, a tool for wiring up and watching simulated brains, which classrooms still use.
1990s
Dial tone.
Before any of that, there was a modem on a phone line I was not supposed to tie up. I found the bulletin boards and freenets before there was a web to speak of, built computers from parts at a shop after school, and took games apart to see how they worked. Everything since has been the same curiosity with better tools.