i'm taking a break

Jul 21, 2026

To do what one wants requires the making of decisions. It also requires the taking of risks. … Keeping the safe job may give you bread and butter and a humdrum existence, with death as the only adventure to which to look forward. Taking the risk may land you in the gutter or elevate you to triumph, but either way there will be a thousand thrills en route.
— In Search of Serenity, R.V.C. Bodley

I decided to leave my job and burn some savings to take a sort of “gap year” or two. A sabbatical, I guess. Right in the middle of a huge shift in the software industry.

After 15 years spent working corporate jobs, I’m ready for a break. I used to find that when I got some time off, I’d spend a week or two away and come back energized. It felt good to get away but also felt good to settle back into the routine, and get some shit done. Lately though, that feeling has shifted. It still feels good to get away… but it’d be nice to stay away. So, I’m going to go ahead and do that.

Leaps are worth it

I’m well prepared for this. We budget, we stick to it, we have enough cushion saved up to last a couple of years. But it still feels like a leap to leave a job with no plan and a hefty mortgage. In my experience, a leap like this is worth it.

I invited my wife to move in with me just a few weeks after we started dating, because otherwise she was going to have to move back in with family halfway across the country. That paid off. It also paid off when we did make the cross-country move together later. Then we moved across country again to be closer to my family, and bought a house.

So here goes another leap. And what a time to do it.

LLMs do not bring me joy

By taking a career break now, I’m stepping out of the industry right in the middle of the biggest upheaval of my career. All the other trends and big shifts are absolutely nothing compared to the shakeup caused by large language models that can generate coherent code. Shifting from on-prem to cloud? A joke. Fully embracing CI and version control as the bare minimum to run a software team? Cute. I can’t think of anything that comes close to the impact this is having on our day to day. And if you buy into the hype around “AI”, it’s only going to get crazier from here.

I’ve tried to have an open mind and embrace it. Naturally, since I’ve just hit 35, my initial reaction is that this is against the natural order of things.1 I’ve tried to consciously avoid that bias and give it a try, but so far my experience has been decidedly mixed.

On the one hand, a fast model in a minimalist coding harness can be a really neat tool. Instead of an IDE having a few specialized types of refactoring commands, you’ve suddenly got a tool you can just ask to make repetitive changes for you. The more boring the change, the better the tool works! I’ve really enjoyed being able to skip some of the boring parts of coding, jot down some comments, and ask the model to fix it. Of course, the flip side of that is sometimes it allows laziness (in a bad way) where instead of taking the time to write reusable code and call it in a few places, you can just have the LLM generate slight variations on the repetitive logic. When it’s just another tool in the toolbox, it’s a pretty normal and useful technology with some pros and cons.

On the other end of the spectrum, we have the whole vibe coding and agentic craze. It’s not just a tool you can reach for; it’s a whole new vibey way of living, man. Suddenly everyone is excited at the idea that they can generate a shitload of code very quickly. Everyone can be a shepherd leading a team of little engineer bots. Personally, it’s not my cup of tea. I got into this because I like programming, not because I like herding cats.

It’s frustrating because agentic development really does feel faster when it comes to getting an overall project done. It takes less time to have the LLM bang out a rough draft and then refine it into shape than it would to work through the code myself. But on a smaller scale, it feels so slow. Now instead of a snappy edit → compile → test cycle, we have to type up a short novel telling the LLM what to do and carefully setting up the right context for it. Then we get to sit and wait for 20 minutes while it spawns a dozen subagents and boils ten gallons of water. Your reward for all this waiting is a big pile of code to read and fix.

We lose the little bursts of creative puzzle solving that come up when you’re actually writing the code. And we trade this creative process for a more boring process of code review. Or, maybe you just accept what the model outputs, and live with crap code.

Will the models keep getting better and start producing great code? I guess we'll see. Will it end up expensive paying a tax to LLM operators to get anything done? Probably.

Doing my own thing

Anyway, I’m going to watch how this plays out from afar. I’m planning to spend my summer fishing, golfing, and thinking about what I’d like to do. Then in the fall, I’m going to do it.


  1. See Douglas Adams’ three rules of technology.↩︎

Changelog

21 Jul 2026 — published initial version