On Claude (and AI)
So I work in big tech. One of the bigger techs. As usual, I'll say: these words are mine and not my employers. I use AI at work. I also use it at home. This site wouldn't exist if it weren't for Claude Code. This is not an endorsement of AI tools, just a statement that I'm using AI to do stuff.
I want to zoom in a little more here. I've had a complicated relationship with AI tools over the past couple of years. They used to suck and I used to laugh at people who used them. The prose was so obviously laden with emdashes, and everything was load-bearing. The code was awful. At work, I'd ask the machine to fix a really simple problem, and it would completely ignore all the tooling and framework and functionality we'd built for solving this kind of problem, and start from scratch.
That's changed a lot. The machine writes better code. I won't say "good" yet, but I think I can say "good enough". So much so that we're now trying to figure out how to manage human review of the large amount of AI-generated code being generated. In the hands of a responsible engineer who tests and manually verifies the code being written, I think AI makes engineers more productive. This doesn't scale infinitely (emdash would go here, but I'm a human without that button so I'll just use a semicolon); I can't just point AI at some very vague problem and have it propose a fix that it autonomously implements. I can point AI at problems and ask it to propose five approaches to fixing it, and I can pick one and have it propose code to implement it, and I can guide the code as it's being written. I can ask the machine to bring the local stack up on my desktop and record a screencast of the fix working, and to run the same test without the fix and demonstrate the break. With supervision, the AI gets me maybe 20% more productivity? I'm writing a reasonable amount of code now, but I also have time to take meetings and talk about other project direction with other engineering leaders. Is this worth the cost? I have no idea and will make no statement as to the real cost. I don't see a mass removal of engineering jobs in the near future, but I do see a future where senior engineers are more productive.
How will we get senior engineers if the tickytack coding is being done by juniors? Not my problem, but boy howdy will that be a problem.
Anyway.
Like I said: claude wrote overstory. I figured it out. I gave the direction, I had the central insights, but I asked claude code to write the actual code for it. As a side effect, it tended to generate prose. One of the first principles docs was claude written. It had the term "load-bearing" in it two or three times. Honestly, when my partner looked at it and said, "did an AI write this?" I was pretty embarrassed, because yes. I'd let a machine speak for me. Which is a bummer, I like saying "load bearing". I guess I write like an AI sometimes. As I get ready to actually launch Overstory, I'm noticing all these claudisms in the interface. A site like this has a ton of user-facing prose, like the text in the archive, or the descriptions of settings on the profile page. I need to replace all of those with actual text I wrote. Sometimes it's just removing some of the verbosity, sometimes it's rewriting an entire passage.

This is my compromise. I will let Claude and Gemini write code for me, because I can read the code and code in itself isn't art. Code can make art, but the art is in the impetus, it is the divine act of creation with intentionality that forms art. Art is in the experience, and you don't see the code. You do, however, see the words. The words are important, even if it's just the making the links to the home say "Go home" instead of "Back to the front door" (which claude really likes).
I'll cop to, and discuss at a later time, how I'm using LLM to actually generate some prose in other places, but in those places it is once again done with intentionality and under my direction (and usually it is by itself the art; the output from the method is just output from the art).
But for now, this is my policy: I write the words.
Conversation happens inside — sign in to Overstory to read and write comments.