Experiments, observations, and ideas: tech tools, AI, books, media, daily systems
Current agent frameworks treat tool definitions as inert data the model can't reason about. The fix isn't a better format, it's a representational property called homoiconicity, and it changes what agents are capable of becoming.…
A ritual from when my kids were small: flip the menu over, draw a grid, fill squares with crayons while waiting for food. I wanted to turn that into an app. Here's what I built instead.…
Today's AI agents can call tools, but ... they cannot evolve them. The bottleneck isn't reasoning or context length. It's representational: actions, traces, and plans live in incompatible formats that prevent agents from inspecting or composing their own capabilities…
Real banger of a quote, by Winston Churchill…
💡The Lamy Safari on the right has been with me since 2012 (and writes just as well today)…
Looks nicer from behind the page!…
Wanted to share Smelter, a self-contained binary for running Lisp scripts without the usual setup overhead. It started as a Coalton runner, but 0.2 adds native Common Lisp mode. Generally meant to solve "just running CL" (i.e. without SBCL images, or configuring Quicklisp).…
I built my daughter an AI tutor that runs on our own infrastructure, uses our own API key, and stores every conversation in a database I control. Open-sourcing it today.…
Worth re-posting from James Clear's newsletter, since it's the only real way to to change: I. "New goals don't deliver new results. New lifestyles do. And a lifestyle is a process,…
Pattern matching isn't just syntax sugar ... it's the compiler asking "have you thought about this?" Learn how Coalton's match expression ensures you handle every case, with no silent nil or missing branches.…
(a different frame for UAPs) I was watching The Age of Disclosure the other night and got stuck at the point where they break down the "five observables": instant acceleration, no sonic boo…
(being an exploration of structural computation) The Premise I've been thinking about how we learn to process collections. In most languages, we reach for loops first – for, while, the familiar…