Why I Joined micro1

I recently joined micro1 as a Member of Technical Staff on the robotics team.

After a few years in the AV industry, I saw the promise of more general robotics and spent the last few months talking to founders, researchers, and engineers across robotics and autonomy. Every team had its own bet: VLMs, WAMs, or agentic robotics; humanoids or arms; sim or real; end-to-end or modular. But almost everyone agreed on one thing. They need more data, and better data, than anyone can produce today. Data remains the most valuable part of developing superintelligence.

I determined that I really wanted to take frontier AI out of the terminal and into the physical world and to help as many companies as possible achieve this vision. While there were many companies providing expert RL environments and data, only micro1 had made a massive early bet on robotics and developed deep connections with experts and businesses as well as frontier labs in AI and robotics.

Not only does micro1 have the benefit of global reach with experts and business, but we recently launched our Malibu data lab specifically for scaling robot training and eval on real world tasks.

All roads lead to robotics

Back in 2024, Eric Jang wrote that once AI systems start acting in the real world, they hit the same problems robotics has been fighting for decades. His takeaway was that “disembodied AGI and robotic AGI happen at roughly the same time.”

Over the past couple of months Astra, Opus, and Argon have shown that advanced models can start acting in the real world. micro1 helped build the data engine for frontier AI to act on real world tasks by partnering with business and experts and now these AI agents are out doing real work. micro1 already sits in the middle of the data loop for the biggest labs, which makes it the natural place to build that same loop for robots.

Robotics are still pre-scaling

While the opportunity in robotics is currently massive, I still believe we are pre-scaling. There are still so many open questions left for us to explore.

That’s what we’re working with AI and robotics labs on building. The robotics team works across the whole stack: from data collection, structuring, and annotation; to research on sim2real, 4D reconstruction, and body and hand pose estimation; to scalable, hyper-realistic sim environments for agentic robotics and RL. We collaborate with researchers on building evals and benchmarks which was further facilitated by our Malibu data lab. This is where labs get real training environments and a place to test their models in the real world.

We’re going past ego video and clean demonstration to figure out which data actually gets robots to superhuman physical skill, across more than one form factor. The current learning paradigms reinforce the jaggedness of frontier intelligence, and as models improve in handling real world tasks the need for complexity and human ingenuity will increase to push capabilities further. We’ve seen this pattern as models went from simple coding diffs all the way to designing entire games and shipping useful services.

The team

micro1 is a tight ship: small, intense, and shipping. Within the first few days I launched our first foray in a new physical AI sector and started collaborating with many customers in bringing their vision to life. The micro1 robotics team is already at $100M ARR in less than a year, and we’re just getting started. If you’re working on robot learning, sim, or embodied data and want to build this with us, reach out. We’re hiring!