AI

Something shifted for me recently. This week, for the first time, I could actually see myself walking away from the flagship corporate AI products. Not from AI itself — from the companies.

This week I did a talk and published a companion website for nonprofit fundraisers on how to get started with AI, including tips, pitfalls, and some of the tools I recommend. It went well, and sitting with all of it afterward, I realized I’d never written a general version of that advice for this site. Most of what I write here assumes you’re already using these tools and looking to go deeper. This post is for those who might be a bit earlier in the journey.

A few weeks ago, Robin Sloan published a newsletter asking a fascinating question: what is it like to be a language model? His answer — that “the model” is the forward pass, a flash of computation lasting milliseconds before dissolving into nothing — sent me sideways into a different question entirely.

Anthropic (the company behind Claude, the AI assistant I use daily) recently released a new model called Claude Mythos Preview. If you’ve heard of it, you’ve probably noticed that it’s not available to us. That’s intentional, and the reason why has me all fired up.

I didn’t decide to take a break from AI. It just kind of happened.

There’s a moment every computer person eventually has, usually more than once, where something gets deleted that can’t come back. Working in the command line for years, I’ve been there. It teaches you, more than anything, that the things you can’t take back deserve a different kind of attention than the things you can.

For over a year now, my favorite large language model has been Anthropic’s Claude. I use it in two ways: the way that’s most familiar to people, chatting in my web browser or their desktop app, and a specialized, programming-focused interface called Claude Code that I run directly in a folder on my computer. Claude Code has actually become my favorite way to access Anthropic models, so when I saw their announcement about Cowork, which is essentially a version of Claude Code that isn’t oriented toward building software, I got genuinely excited. That’s exactly how I’ve been using it myself.

I recently heard Cory Doctorow speak about enshittification, his framework for understanding how digital platforms decay over time. The talk was wonderful and thought-provoking in the way good author talks should be. Cory challenges my thinking regularly, particularly when we don’t see eye to eye. I actually agree with most of what he says about AI, though my personal experience leads me to draw different conclusions from time to time.

I’m getting nervous about my new web development class now that it’s just a few weeks away. It’s my first time teaching older students — tenth, eleventh, and twelfth graders — for a full semester, and I’m doing it alone. Last year I dipped my toes in with a smaller group, learned some things, and made some mistakes. But this is different.

Last year, my students were hitting usage limits on free ChatGPT accounts mid-conversation. I was working hard on prompts to help them use AI as a thought partner rather than a shortcut or search engine, but I kept running into limitations.