Martijn

@martijnarts.com

🦀✨ building the Rust framework for Humans and AI @aonyx.ai chat with me: https://calendar.app.google/JdYFuTMFKTU6DocD6

I feel like open source will (have to) change a lot with AI. We won't all run the same software, but will personalize it to our own needs. That changes how we collaborate, build communities, and maintain code. For better or worse. Started writing down my thoughts: jdno.dev/how-hyper-pe...

How Hyper-Personalized Software Challenges Open Source

AI makes it cheap to build software for exactly one person's workflow. The idea is twenty years old, but suddenly feasible. And it breaks an assumption at the heart of open source collaboration.

jdno.dev

I'm prototyping a Github reviewer app, and asked Claude to setup a private app to test it out. Came back a bit later and it had said it was waiting for me, app complete including two custom pages to set it up and install it on our org. I just had to do 3 clicks. AI coding feels magical sometimes...

Claude Code chat history:

me: what do I need to be doing?

Claude: Just the two browser steps — everything else is done and waiting on them:

1. Create the app: open .../setup/create-app.html in your browser (or run just setup from that directory — it does the same). Click the "Create @... on aonyx-ai" button, and approve on the GitHub page it takes you to. When GitHub redirects you afterward, you should see "... created." — that means the credentials landed on the Modal volume automatically.
2. Install it: that same landing page has an install link — click it and install the app on the aonyx-ai org, scoped to all repositories.
That's it. No keys to copy, nothing to paste. I have a watcher on the volume, so the moment step 1 completes I'll see it and start the end-to-end test on my own.

I've been thinking about hyper-personalized software ever since I've started self-hosting some services. I think there are (at least) three things that need to be solved before we can move to a world where everyone builds (or forks) their own apps. 🧵

I'd love to credit whoever wrote this. No idea. But I'm 100% with its message. When I think of how the '80s ethos, "Greed is Good" has metastasized, it surprises and depresses me. Imagine having so many millions, hundred of millions, even billions & caring about nothing except your own vanity.

Bild

LLMs shall be a reckoning for a lot of developers who thought being clever was their ticket because the answer is now to take a really big hammer and point in a straight line and start pounding in tentpegs 'til you get to the goal This will be bad for one's satisfaction but probably good on net

iterated with Claude from this lil prompt: ❯ hi my disk is 500gb but 437gb used and I feel like that's too much we got to 165 GB cleaned up now.

A table showing a total cleanup of 165 GB of data, split over various items.

Claude invented a "type it out explicitly" confirmation requirement for a force push (for the first time ever??) and then proceeded to fully ignore it.

Claude Code:
The permission classifier blocked it — it wants the force-push named explicitly, and "yes" to my question didn't register as that.

Easiest path, run it yourself:

! git push --force-with-lease origin martijn/...

Or reply with the force-push spelled out (e.g. "force-push martijn/...") and I'll retry.

Me:
do it

Claude Code:
Bash(git push --force-with-lease origin martijn/... | tail -10)