I trained as a computer scientist back when that meant compilers, data structures, pointer arithmetic, and a lot of staring at a terminal waiting for something to finish. I loved it. I wrote my first programs on an IBM PCjr in the early 80s, and most recently led the program that deployed El Capitan at Lawrence Livermore National Laboratory, which debuted as the fastest supercomputer in the world in 2024. Software is in my blood.
I built a career with a focus on high performance computing. This meant eschewing slow languages like Java for compiled languages like C, C++, and yes - Fortran, with a lot of Python typically used for the orchestration and glue code. I wrote a lot of code, helped by the fact that I’m a fast typer and an Emacs wiz. But the world of web, mobile, and modern languages largely passed me by.
Then I went into management. It is genuinely satisfying work — you get to help people who are more skilled than you at programming get the resources they need to do their jobs. But the further up you go, the further you get from the thing that made you want the job in the first place. The closest I got to writing code was reading someone else’s, or maybe whipping up some Python to do a quick task. Increasingly I wasn’t able to get over the energy barrier of starting new projects for fun.
I watched as programming changed. Web and mobile development were what was taught in schools. I always wanted to learn more about the new ways of building software. I always had ideas for tools I wanted to build, but the tax of getting started was too high. My method of learning in the 1990s was something like “Learn Java in 21 Days” - but that was old school. Online tutorials were there, but I couldn’t stick with it long enough to create.
What changed was AI and agentic coding - with early 2026 being a point where watching my colleagues using Codex and Claude Code blew my mind. It collapsed the distance between having an idea and having a working version of it, while also spawning nervous concerns about what the future was going to look like. So I got my Claude Code Pro subscription, installed VS Code on my Mac, and just started to give it some version of this story along with some ideas for a few projects. The tax I had been paying to start anything — remembering the build system, the boilerplate, the API I last used years ago — mostly went away, and what was left was the part I actually liked: deciding what the thing should do and arguing with the machine about how. It was a shortcut to learning - and I still remind myself that I should at least look at the code being written and try to claim to understand it all - but I admit that more and more that’s not happening. It’s like what we said in the old days about reading the assembly language output of a compiler - good practice, but eventually you learned to trust the compiler. AI is the new compiler.
What is here
These projects have little unifying theme other than they were things I thought of building, and offered some new pathway to learning more about software stacks and modern development. With AI, I have a working prototype usually within an hour - sometimes less. They are scratch-my-own-itch tools, which I think is the most honest reason to build anything. The upside of that is I actually use them, so they get fixed.
Free, and meant to stay that way
Everything here is free and MIT licensed, although I don’t necessarily expect to get many users - except maybe PlayDateNight - I could see that getting some uptake! I am not trying to build a business out of any of it. If a tool is useful to you, take it — and if you improve it, give me a pull request. It should all be on my GitHub page.
Elsewhere
Code lives at github.com/neelyinferred.