Now what?
19 points by nathell
19 points by nathell
Similar to many developers in the pre-LLM era, I had a decent amount of unfinished projects and ideas that I thought it would be cool to bring to live. Most of them I gave up midway because, once I started to work on it, I didn't have the time, energy, or it was just not such a great idea after all.
Nowadays I have access to the yes machine that can produce these projects for me and I still don't do it. Why? Honestly, I feel the human bandwidth is a good proxy to tell if something is important enough for you to build, care and ultimately use it. I know people vibecoding their super tailored apps that replaces their SaaS subscriptions makes a good headline, but this is just survivorship-bias. I believe the majority of people that finally concluded their software dream project because of AI, have already found out that finishing was not the problem.
I hear people say this and am baffled. My side projects I have always either built knowing that I was not going to use them or because they filled a need (or want, or very minor desire) I actually had. This has not changed with LLMs.
To pick an example at random, I ported facebook's Demucs stem splitter to run in wasm, because I wanted something I could use on my phone, and I've used it several times now to give demos to kids about listening to various parts of music. Could I have done that on my own? Absolutely. Would I have? Empirically, no.
My wife has way more of these than I do - e.g. a metronome app for running that works exactly how she wanted it to, or a pattern designer for weaving. She also uses hers all the time. Prior to Claude Code she had mostly stopped programming (she did it professionally but not as a hobby).
I cannot imagine looking out at the world with a programmer's brain and not seeing a bunch of things which could be at least minorly improved by application of some code.
I've realized that most of my pleasure in side projects came from the act of building, not the thing to use afterwards. That's probably also why I abandoned so many of them when the pleasure of building declined below the burden of maintaining.
I’ve begun the process of rebranding myself from “software developer” to “software craftsman”.
Very similar experience here. I always have a lot of ideas I want to try doing, little things I want to automate, or just to see if something works or not. Before LLMs, the barrier to figuring out all the boilerplate, library APIs, tooling, and so on was just too high to bother unless you really wanted to make something work. Now, it's really easy to just get the LLM to build a small tool that scratches an itch, and if it doesn't work out then it's not a big deal either.
I've resurrected a whole bunch of half finished projects I've had, and got to explore a lot of things I just never had the energy to do before. And a lot of the time you do learn something new. Perhaps not at the same level of detail as you would have if you did it artisanally by hand, but you still learn something in the process.
For example, my latest adventure was to try and reverse engineer NEF files my Nikon camera uses which don't have an open source decoder. I use Topaz Photo to clean up noise and sharpen photos, and I wanted to see if I could use an open source model to do that instead. In the process I learned how wavelet based codecs work, and that gave me an idea to apply wavelets to code analysis. It turns out nobody tried doing this before, and it resulted in a really neat tool I now regularly use to find gnarly parts in codebases https://yogthos.net/posts/2026-06-02-wavescope.html
So, I got two useful tools out of this experiment, and I learned about a technique I didn't know about in the process. This would've never happened without LLMs because I wouldn't even have bothered trying to reverse engineer a proprietary image encoding format on my own.
I cannot imagine looking out at the world with a programmer's brain and not seeing a bunch of things which could be at least minorly improved by application of some code.
Let me be the first to congratulate you on learning that people do not share the same priorities as you.
To be clear, I'm absolutely not saying that any of this is or should be a priority! There are also lots of things that could be improved by application of money, or time, or whatever, and in almost all cases it wouldn't be worth doing.
Exactly this. LLMs removes constraints around time and energy that in my opinion help produce better outcomes.
for a more playful example, the one-shot editor for poems in Pilish was posted as an aid to a Facebook group
Clearly off topic, but thank you for making me discover Pilish. Made my day honestly. Now a colleague has challenged me to write the next PR description in it.
Sorry for the off topic comment, but your blog is absolutely beautiful. Everything about the styling / font formatting is spot on. Feels like reading a book. Did you use any style guidelines or rubrics to set this up?