What parts of my software should be tested? And how? And how do coding agents (e.g. Claude Code) change things? This is my attempt to succinctly explain how I think about these questions at the beginning of 2026 (software development is changing so quickly I feel compelled to note the date. This may all be obsolete soon).
In this post I will briefly describe Earl, a reinforcement learning (RL) framework I wrote that enables scalable distributed training across multiple devices, and discuss some of the things I learned along the way.
In this post I describe some things I learned while working on Starflate, an implementation of Deflate decompression in C++23 that I wrote with my friend Oliver Lee.
While working on machine learning research at the Astera Institute1, I led a team that assembled a system that enabled researchers to quickly and easily run experiments that used up to a full datacenter’s worth of GPUs. I intentionally wrote “assemble” rather than “build”, because the system mostly consists of off-the-shelf components. The challenge was in digging through the huge number of…
This post is written as a Jupyter notebook which you can run and edit using the link below:
You can use an account at Fidelity as a checking account, meaning you can write and deposit checks and withdraw cash from ATMs. Why do this?
Kagi is a relatively new search engine. Unlike Google, it makes money through user subscriptions and shows no ads. Despite having decreased my usage of web search since the release of ChatGPT, I still use it a lot, and would be willing to pay a few bucks a month for a significantly better experience. To evaluate Kagi, I put 75 of my recent search queries into Kagi and Google and rated which I…
Here I present some ideas for increasing the productivity of knowledge workers by structuring their workflows around attention and memory. I wrote this for my own benefit, but I hope you find it useful too!
In The dangers of reproducing while old, I mentioned that pre-implantation genetic testing could be a way for older prospective parents to improve outcomes. This led my partner and I to seriously consider doing IVF despite not having any fertility issues. At a high level, my conclusion after writing that post was:
I had my first child when I was 36 years old, which made me want to understand the risks of having children at different ages. Before looking into this, my impression was that the main biological problems with old-age parenthood had to do with not having the necessary health and vigor to care for young’uns, and I had heard that older women have trouble getting pregnant. While those are real…