Frisky and Xarray Example
Let's run Xarray workloads on this new thing I built, Frisky, a Dask scheduler in Rust.
Personal site of Matthew Rocklin - thoughts on open source, startups, and engineering
Let's run Xarray workloads on this new thing I built, Frisky, a Dask scheduler in Rust.
Six months into AI-assisted work, the biggest gains come from broad context, cross-domain reasoning, and feedback systems that help agents converge.
I'm stepping back from Coiled, Dask, and Python OSS to make space for what's next.
AI makes software easy to build, eroding the winner-take-all dynamics that created tech giants. The future may look less like Google and more like clothing brands
A week of plan mode, custom themes, performance optimizations, and the --yolo flag
Exploring multi-agent AI workflows—what works, what doesn't, and lessons from playing Diplomacy with Claude agents.
Everyone feels behind on AI, but no one is ahead. That shared beginner status is what makes this moment so fun.
A Claude code drop-in with updated UX and multi-agent support
Senior engineers are best positioned to benefit from AI. We're good enough to avoid slop, and there's so much we can accomplish. I wouldn't go back.
LinkedIn shows only the final act. We connect to heroes during their struggles, not their victories. The solution: fail heroically in public.
Reflections from 27 days and 500 miles across Spain. The overlapping physical, social, cultural, and spiritual layers that make this pilgrimage meaningfully different.
Speaking flows at the speed of thought. Writing forces you to organize first. This is hard on you but easy on your reader.
"Can I ask for your help?" makes me feel like a person who helps people. "I'll pay you back" makes me feel like I'm trading favors. Incentives feel transactional.
A nailgun puts nails into wood. A hammer whacks things. Simple tools with humble opinions are more versatile than specialized ones.
When a Python workflow fails, most of the work is diagnosing whose fault it is. The ecosystem's distributed ownership means no one has clear responsibility.
What hunger am I seeing on the ground? Boring shit. People struggle most with the basics of cloud computing. The promise remains undelivered.
Constructive conversation and venting differ by one question: "Will these words help me achieve my goal?" If no, we're probably venting.
Satellite imagery's potential is incredible, but companies in this space are struggling. An outsider's first impressions from an unconference in Boulder.
When a small amount of your work helps others a great deal, of course you sacrifice. You'd be a monster not to. But your ego can fade in service of your project.
In 2018 we unveiled a new architecture for scalable cloud Python. It was great, but also fragile and limited. What comes next?