Have spent all of 2026 building something new post-FaZe. Excited to join @ycombinator this summer as I continue on this next chapter.
Ten years ago I was 20, moving out of the FaZe House in New York to our new place in Newport Beach.
This week I turned 30, and Iβm feeling the same excitement here in SF that I felt in 2016. Excited to share more soon ππ»
AI is going to have a hugely positive impact on allocating human intelligence to build really cool technology like this.
Many folks I know for years have expressed negative sentiment for the tech industry along the lines of βWe thought we would have flying cars by now, but instead we have X boring <insert B2B SaaS>β
But with agents and models working together to commoditize intelligence and much of the application layer, more of the smartest people in the world will work on things like flying cars. Or drug discovery. Or space exploration.
The sci-fi things nobody thought could be possible. Very cool from Midjourney.
Our autonomous harness engineering system is being tested & deployed with frontier AI teams you know.
@mayonkeyy & I have spent hundreds of hours locked in a room thinking about how to make it. And we're excited to share our learnings & a guide on how to start building it with all of you today!
In a decade, there will be a very clear SpaceX Mafia.
Iβm lucky enough to know several very smart folks on the team that made a lot of money today.
The culture, particularly in MAE at SpaceX is more mission-driven than any company of that size I know of, and it flows through the entire team.
Great for the world that it gets rewarded.
Statistics needs way more emphasis in school systems.
AP Stats was easily the most useful class I took in HS (one of the few), substantially more so than any Calculus past basic derivatives, and has applications in basically every domain.
Agent benchmarks will soon shift to online outcomes, just like humans.
The Midas list looks at investor performance, not exactly how an investor does diligence / sources.
Boxers are ranked based off of how many wins they have, not exactly how they punch.
While humans can still define βgoodβ and reliability is still difficult to achieve for complex tasks, offline evals are critical.
But after, how do agents get better based on dynamic, real-world outcomes? How do they experiment on themselves in prod?
If a product onboarding agent is tied to product activation, what % of users did it activate?
Thatβs what weβre building for now.