Why I joined micro1

I've joined micro1 as VP of Engineering. A few people have asked why, so here's how I got here.

I spend a lot of my time talking with founders, investors, and people building AI systems. Over the past year, one theme kept coming up: progress on frontier models is increasingly limited by two inputs, compute and data.

Compute gets most of the attention. Because of my background, I paid closer attention to data. I talked with many people working at data labs and on the teams that buy from them. I wanted to understand what labs are paying for, where quality breaks down, and how fast needs are changing.

The conclusion was clearer than I expected. Demand for high-quality human data is growing much faster than the supply of people and systems that can produce it. This demand isn't for generic labeling. Labs need domain experts who can write hard tasks, judge subtle answers, build realistic training environments, and explain why one response is better than another.

I also took seriously the strongest counterargument: this demand is temporary, and once models are smart enough, they won't need human data anymore. Ali's case against this convinced me. Even self-play and synthetic data rely on humans to define objectives and what "good" looks like. Automation also happens gradually, one function at a time, and each step needs new demonstrations, evaluations, and corrections. And as automation frees up human time, people create new kinds of work, which then go through the same cycle. On that view, human data is an ongoing input to how AI improves, not a temporary bottleneck. He lays out the full argument here.

So the market made sense to me. What made micro1 specifically the right place was the people.

"Human first" is micro1's most important value, and it isn't a slogan. Ali's goal is to build the most human-centric company on earth. In practice, that means the experts who do the work are treated with real respect, and they like working here. For example, more than 1,000 lawyers work on micro1, earning on average about 20% more than in traditional firm roles. That respect shows up in the quality of what experts deliver. Labs notice that quality and send more work. I think this loop is a big part of why micro1 is growing as fast as it is.

My engineering job is to strengthen that loop. That means building tools that let experts spend their time on judgment instead of on the machinery around it. It means infrastructure for reproducing evaluations, tracing dataset history, and catching quality regressions early. And it means an engineering team that moves fast while staying rigorous about evidence.

At Apollo, I learned a lot about scaling an engineering org and shipping AI products people depend on. One of the most useful lessons was how much domain experts contribute. Engineers alone often can't tell whether an output is actually useful. That lesson applies directly here.

Growth like this doesn't come around often, and I'm glad to be part of it. If you're an engineer interested in how AI acquires real expertise and how we verify it, I'd love to talk.