Trigger warning: AI hype and worries. sorry everyone

It’s been about 9 months since I’ve been Claude Coding, and 6 since it’s been my primary mode of coding. That is, “open a terminal, tell Claude what I want to do, let it whirr away for minutes.” In early 2025 I was using Cursor to pretty good effect, writing my PRs and asking it for help on parts of it. And if Cursor was all we got, it’d still be a productivity doubler or more. But Claude Code has made coding qualitatively different. This has given me lots of feelings. In this post, I try to sort them out.

first, my preferences

Here’s what I like about making software:

1. Puzzle-solving: “I need this part of the app to do x. But that’s tricky; how do I do it and make it fast?”
2. Creating: wanting a thing, then seeing it exist in the world
3. Organizing: closing all the tickets, organizing all the files, putting all the bits and bytes exactly where they should go
4. Learning: how to make new things. Whether it’s about my current project or software in general: “how have these tools, that I’ve been using/importing/depending on, been working all these years?”

personally, this rules

It is so much more fun to work on side projects these days. Side projects always have a balance between A. the excitement of having a thing in the world that you want to exist, and B. the pain having to look up API documentation and all the other grunt work. If A > B, the project happens, but usually B > A and it doesn’t. Nowadays, B is almost zero.

I hacked together Closeness To Things in a small number of hours. Usually it would take me that many hours just to figure out what geocoders exist, try to compare them, realize there is no one best one that fits all my needs, and get frustrated and quit. Claude just implemented four different ones for me! And three routing engines! And maps and UI buttons and all the annoying little bits. The People loved it.¹ I loved it! I have the satisfaction of website out in the world, almost for free!

So, lots more creating joy; that dopamine hit in a day, not a month or year. Puzzle-solving joy has dropped but it wasn’t very high in the first place; when I start a project from scratch, usually the challenges aren’t very interesting. I don’t usually start projects to “improve Traveling Salesman bounds”; usually it’s more like “I wish there was a website that does X.” (Often I will get to an interesting problem, but sometimes all the overhead of setting up tools kills the project before then.

Organizing joy is a mixed bag: LLMs create so much more stuff that it makes me a little twitchy. But they also help me organize it; I recently did some blog migration that went fantastically well and easily. “Find the pictures, put them in this folder, sorted by year, you know, the smart way.” If a task is something I could delegate a high schooler to do, LLMs obviously nail it.

Learning joy also way up. It’s shallower and broader; now I have a sense of the geocoding and routing ecosystem while still not knowing it in detail or being able to implement a geocoder. But that’s something; before this project, I didn’t even have that; without LLMs, I probably would have gotten hung up on the geocoding before ever getting to the routing.

professionally: huh

I think it’s a blessing, but it’s more complicated. Why?

do more business, faster

Imagine you work at Toyota, and your engine supplier ToyEngineCo announces that they’ve created a Double-Good Engine. It’s twice as fast, twice as efficient, twice as everything.² Is your life better?

Not really. Pretty soon ToyEngineCo is going to sell Double-Good engines to Honda and Ford, or else HonEngineCo and ForEngineCo are going to figure the technology out as well. Then you’re in the same race you were last year, but you’re both putting out products that are twice as good. You can’t start working 20 hour weeks just because you’ve got Double-Good Engines. Your stock options probably go up in the short term, but not in the long term, because everyone will have Double-Good Engines.

Not that I’m trying to do less work; just, you’d think 10x-ing your productivity would make you super rich or successful or something. It doesn’t, because everyone else has 10x-ed too! And 10x-ing everything means you have 10x more files, 10x more features, 10x more projects to keep track of. Organizational joy way down.

I think creative joy has gone up, but it’s hard to say, because professional work never ends. We make a data warehouse; now it’s 10x bigger and better than it would have been without LLMs, but we don’t ever see the one we would have made, so it doesn’t feel like we created a “10x bigger” data warehouse, it just feels like we made the thing we wanted to make, in the time we took. Sometimes I do think about past projects and realize that I’m creating an equivalent much faster, so I feel some extra creative joy, though, so that’s some perspective. It’s not as dramatic as personal projects, though.

Claude’s eaten the meat out of the sandwich

The Decide-Execute-Deliver sandwich, from AI as Normal Technology.

Software involves these 3 tasks: deciding what to do, writing the code, and then delivering it (testing, deploying, maintaining). The middle part costs maybe 10% of what it used to. That’s great if you like the Deciding and Delivering parts, and kind of a bummer if you like the Execute parts.

I did like the Execute parts. I liked writing the code. I certainly can do the Deciding and Delivering, but the Executing was the most fun. I didn’t like writing all the code, so I’m glad to miss the bits where you had to look stuff up, but the parts where you were in flow, edit-test-debugging, that was why I got into software engineering. They’re not completely gone, and they’re often more fun (“Claude move it over there, figure out the CSS” instead of doing it myself), but there’s much less of them. Puzzle-solving joy way down.

upskill quickly

Now, when I’m trying to learn a whole new thing, I have the best learning environment: an expert I can ask, ad nauseam! In this job over the last half year-ish I’ve gone from b2c tech and internal apps to b2b data warehouse platform, and there’s no way I could have learned all that as fast as I have without LLMs. So, learning joy way up.

putting numbers on it

“How much I feel these joys” on an arbitrary 1-10 scale:

I don’t know. The “organizing” hit is real and merging another +3000 PR without thoroughly understanding every bit of code, or pointing an agent at a whole Linear board and saying “let ’er rip” feels scary every day. I feel frazzled by the amount of context switching required by LLM life, and I don’t think we’re ever going back. I wish I had the type of mind that worked well in that kind of “ER doctor” environment.

But I do have more skills, less drudgery and a lot more possibilities, and tbh the time spent programming is probably less frustrating all in all. Problem-solving joy is gone, but creative joy is still there, so I suppose the way to flourish now is to focus on that as much as I can.