A sheepherder's son finds a computer

In the 1990s, my parents saw Bill Gates say on television that his goal was a computer on a desk in every home. Not long after that, they bought a hand-built 286 from a friend and put it on a desk in our house. It ran MS-DOS and had almost nothing installed on it. I would open the command prompt and type natural-language commands such as let me write a paper or how are you doing? The computer always answered the same way: command not found.

My father immigrated to the United States from the Basque Country in Spain as a young man, where he worked as a sheepherder. He later met my mother, who had also immigrated from Spain, and together they raised our family in Vacaville, California. They were both blue-collar workers, and neither of them knew what to do with the machine, so it mostly sat idle. I was fascinated by it, but I had no frame of reference for how to use it or even how to learn. I opened the case and stared at the technology inside, mesmerized by a machine I could not yet understand.

That changed when one of my mom's coworkers came over. She was the first person I ever saw actually use the computer. She installed software she had purchased herself, including WordPerfect and several games. In her bag, I spotted a game called Leather Goddesses of Phobos. It was an Infocom text adventure, although I did not know that at the time. I was around fourteen years old, and the title sounded incredibly dirty, so naturally I had to play it.

My mom's friend noticed my interest and gave me a challenge.

Johnny, I'm going to hide this game on the computer. If you can find it, you can play it.

She also left behind a copy of DOS for Dummies. My mission was on.

I read that book page by page. There was no internet available to me, and it was the first real instruction I had ever received on how to use a computer. Eventually, I found the game. I was victorious, even though it turned out to be much less dirty than its name suggested. That challenge started a lifelong fascination, and what became one of my deepest special interests: computers.

I bought a book about programming in BASIC and began writing my own software. Some of my earliest programs were launchers that let me type commands in something closer to natural language, much like I had tried to do the first time I sat in front of that MS-DOS prompt. Later, I taught myself Perl. I dialed into BBSs with a modem. I tried to learn everything I could about this wonderful machine and the world it opened up.

When I finally got an internet account, I quickly found ways to get into trouble with it. I discovered that I could crash Microsoft Comic Chat clients using CTCP floods. What I did not realize was that I was also crashing the servers behind them. Microsoft was not very happy about that, and my internet account was shut down.

That was also when I began finding communities I had never been able to find where I grew up. I found 2600, other teenage mischief-makers, and the San Francisco rave scene. The internet connected me to people who shared my curiosity, my interests, and my appetite for exploring systems that were not always meant to be explored.

When I get lucky and my hobby becomes a job

At around sixteen, I got my first programming job from a man who ran a stamp-collecting website. He paid me minimum wage by the hour to update the site and work on its checkout process. When I was eighteen, I got my first job at a dot-com, Rentals.com, about an hour and a half from where I grew up. I moved to Silicon Valley for the first time. Around that same time, I met a girl at a party who was later able to find me online. Five years later, she became my wife.

After Rentals.com and several other startups, I joined Tellme Networks. Before mobile speech applications were common, I traveled around the country working with major telephone carriers to automate 411 directory assistance. Later I worked on what was among the first speech recognition applications for the BlackBerry platform, in the mid-2000s. Before smartphones made voice assistants commonplace, we were putting speech recognition into people's pockets and connecting them to large-scale voice services. Microsoft eventually acquired Tellme, which was a little ironic considering my earlier history with Comic Chat.

At Microsoft, I worked on speech experiences across a wide range of products, including Bing Voice, Xbox One, and several Windows speech features. Later I helped launch the OpenAI Realtime API on Azure, which was one of the most exciting projects I have worked on because it represented a genuine shift in how speech systems are built. It was the first production speech-to-speech model of its kind, and the Realtime API made it possible to use. Overnight, it made many of the approaches I had spent years building feel obsolete.

What made that project especially meaningful to me was that it brought together so many different parts of my background. I had spent years working in speech recognition, telephony, networking, distributed systems, and low-level software. I also still had the curiosity of the teenager who loved taking computers apart to understand how they worked. Shipping the service on Azure alongside our partner required all of those skills at once.

Today I work on Azure AI Foundry. While I still work on APIs, much of my time is focused on using AI agents to improve how software is developed, tested, and delivered. I am particularly interested in building secure environments where agents can work independently without creating unnecessary risk. That interest naturally extends into sandboxing, virtualization, and desktop architecture, which has become a major focus of my personal projects.

My current fascination is large language models. I have been waiting a long time for computing to reach this point, where a person can finally interact with a machine in the way I instinctively tried to when I first sat down at that 286.

Elsewhere

My most significant open source project is D2B, short for Double Dutch Bus. It is my attempt to build the desktop environment I have wanted for years: one that makes it easy to work across multiple trust boundaries while safely collaborating with AI agents. It combines virtualization, secure brokers, Wayland integration, and a Kubernetes-inspired control plane into a single desktop experience. The goal is to reduce cognitive load by making secure isolation feel natural instead of getting in the way.

D2B and the rest of my code live onGitHub (opens in a new tab).

Privacy

This site uses Cloudflare Web Analytics for aggregate traffic and performance measurements. It does not use analytics cookies or track visitors across sites.