on ai

this will be a bit of how I view and use ai, as of July 2025.

firstly, if I’m reading something and I realize it was written by ai, I lose all interest. it’s actually kinda crazy. the corollary is that I will not use ai to write anything important for me. any Writing. if I want to be the Author, no ai.

secondly, I use ai a lot, mostly for programming. my favorite term for writing software with ai is “context engineering”. I provide the machine with the necessary context to understand the problem, and then it solves it for me. this use case also applies to any professional/bureaucratic writing that I need to do, or approaching a problem outside my areas of expertise.

what ai do I use?

as of right now, I pay for claude, and use perplexity for search. I don’t use any of the fancy tools to write code (other than supermaven), but I’m interested in something like aider, I just haven’t put effort into learning how to use it. (note to self: check out this video to learn more about the topic) I think when the year is up, I will likely switch to t3 chat (so that I can have a larger variety of models to try) or something where I can use an API so I can try out aider. when searching for something, if it’s a fact/link I want, duckduckgo, if it’s a question, perplexity, and if it’s something I only tangentially remember, then google is still good at that. but if I know specific search terms, duckduckgo works well.

how do I use ai, specifically?

ai is very good at getting me started on things which I know nothing about. the other task which ai is very helpful for is reading boring material, and getting answers out of it. contracts, government laws, tedious emails, documentation, that sort of thing. this combine when I give the ai documentation and an example, and then ask it to do what I want. for example my kanata keymap was partially written by ai. I didn’t know the syntax, but I gave it my QMK keymap, an example in kanata, and the documentation, and asked it to do the same thing with kanata. (actually now that I think about it I wrote the first couple drafts, but used it to help me with more difficult features.)

what is ai not helpful for?

at the same time, if the problem is novel or obscure, it is a big waste of time to use ai to help me with it. I have run into this a few times. for example, I wanted to write code to essentially do a fit diff between two text strings. it wasn’t helpful.

sidenote

today I saw a tweet about a study where they showed that software engineers were slower at writing code using ai than they were without it. that’s crazy to me. the engineers thought they were 20% faster and were actually 20% slower. I am sure that more people will study this, given how much money is wrapped up in software engineering. it looks like a big part of the slowdown was from waiting for the output to generate, and context switching in and out of the problem during that time.