AKA Please, God, Not Another AI Thinkpiece!
Nope. There have been enough of those lately. Recent posts about art, intuition, semantic ablation, cognitive debt, cognitive shortcuts and atrophy. They get at lots of the nuance hidden between “AI is literally SATAN” and “I, for one, welcome our new AI overlords!”. Mostly, (generative)AI is kinda useful for some things, is extremely problematic for many reasons, and isn’t going away no matter how much anyone wants it to.
I’ve been using Claude1 a lot - Claude-Coding up a storm, building lots of little shiny trinkets and tools. Several of these, I use daily. My most-used GenAI-built thing is my bookmarks application2, but there are a few other daily drivers. And I have a whole collection of custom Obsidian plugins3 and standalone scripts and applications that aren’t even listed there. Claude also automatically generates tags, descriptions, and alt-text for all of the photos that I post here. I also use Claude for summarizing things, converting handwritten notes into markdown, etc.
Screenshot of Claude Code in the macOS Terminal application
The stats only shows usage since migrating to our Claude Teams accounts and after Claude enabled Code for all Teams accounts.
Image by D'Arcy Norman
The revelation I’ve had recently is that I use genAI for making things that I don’t “care” about. The web-tools? I care about the thing that is made (sometimes) but never about the code itself. The summaries are for things that I didn’t have the energy to read in the first place and only needed a passing understanding of what a thing was about - not a deep, close reading, but more of a “holy hell this is a 20 page article that should be maybe 3 paragraphs - see what you can pull out of it” kind of thing.
So, for those web-tools, it’s not about the code as much as it’s about just willing something into existence in between (during) meetings to see if it’s useful. Most of them get used a bit and then discarded. A few get used every day. And I didn’t have to stop and learn 2026 web design frameworks and new CSS (and and and) in order to get something that worked. The usefulness was in having a thing, not in the building of it.
And, yes, of course, technical debt and cognitive debt and shortcuts, but also art and intuition and playful creation that wouldn’t have taken place without the tedious work being handled by genAI tools.
I won’t use Claude to create something that I want to actually learn or for something where the entire process is important to me. I think it’s potentially useful to be able to decide that I don’t need to create the universe to make an apple pie, because that means I can reallocate energy to let me care more deeply about the things that are more important to me, and to my work. (DLN: ok. that last bit was overblown. I’m writing about making web trinkets and then trying to connect to Deeply Caring, which, sure, but that’s quite the leap. anyway…)
our team is piloting a “Claude Teams” license for a year to see how it might be used - having access to a genuine “frontier model” is pretty important when we’re trying to have informed discussions with instructors and university leadership about what generative AI can do and what the implications are, and Copilot just isn’t there. ↩︎
A pretty feature-complete custom-built drop-in replacement for Del.icio.us etc, but for one user. My live Bookmarks application is at https://darcynorman.net/bookmarks/ and the source code is at https://github.com/dlnorman/standalone-bookmarks ↩︎
Including an LLM “related notes” plugin that uses vector math to find notes that are similar to the current note without having to rely on explicit tags and links, a “People” and “Topics” interface that make my notes extremely useful and efficient, etc. ↩︎