The things I never wrote about
Published .
I have always been sharing my thoughts with the people around me, ask them. And, from time to time, I even enjoy writing them down. I honestly have no idea why I have not gotten to this sooner. At some point, I started sharing some of my thoughts on Mastodon. I really like microblogging, but I use it most of the time to either share random thoughts or to leave comments on other peoples posts. When I actually think of something interesting that I want to share with the world, five hundred characters will not do the trick. So I am starting this, feel free to follow along for the ride. Mastodon will maybe see a note when I post here, and so will the Hacker News front page, at some point. One can have a dream, right?
Anyways, my “Note to Self” chat in Signal is full of ideas that I have thoughts on, things that I noticed at work or in private, and other blog posts that I found really interesting. This post will just be a random collection of things from that chat, do not expect any coherence between any of the following sections.
If you want to learn a few things about me, have a look at torge.rdahl.de. If you want to get to know me, leave me a message somewhere. I will not do any personal introduction for now, this blog is meant as a channel for my thoughts, not for me as a person.
No AI is or will be involved in this or any the following blog posts. If you spot a typo, good for you, I don’t care. If you spot a factual error or something incomprehensible, let me know.
A federated Software Forge
Github has turned from a really helpful platform into an honest risk for open source software and freedom. GitLab is nice, all the companies I have worked with so far have had at least one GitLab instance, usually self-hosted. Having an open source platform to share open source code is the only way. But GitLab is also pushing very hard in the direction of AI, meaning that it is necessarily moving more towards a centralized service. Then, there are new stars on the horizon, the brightest one in my hemisphere seems to be Codeberg / Forgejo. And there is a broad range of other, good and bad, platforms that also play their important roles!
Those forges are all running Git. Moving a repository from Github to Codeberg is quite easy, it took me maybe half an hour total for the maybe 20 active repos that I moved two years ago. That is definitely a good thing. But what it also means is that source code for very nice and important projects is scattered all over the place. The Github feed used to be really nice to follow the development and release cycles of interesting projects, but it has become a lot less useful with more and more projects moving to other forges.
Two ideas I have on this: Either, you federate together those feeds, which might actually possible through a technology as simple as RSS. Have not looked into it, probably will not look into it. Or, you build a full on federated suite of software forges.
How amazing would it be if you could follow users and collaborate with users from other instances of Forgejo, for example? Nobody is forced to stick with Codeberg, you could self-host, but people that are still on Codeberg (or any other platform) could star your repository and receive updates on it, maybe even contribute to that repo without creating a 10th account? If we throw ActivityPub into the mix, maybe it could even be possible to follow discussions or the latest releases from a platform other than this specific Git forge?
Yeah, anyways, I am not willing to build a proof-of-concept, much less a working version of this. I think there is a feature request for this at Forgejo, if you build it, credit them, not me.
A portal for ideas I have had but did not act upon.
I once wanted to create a portal where I could post random ideas for helpful pieces of software and other people could basically “pick” those and then act upon them.
I did not do it and in hindsight, I think it would not have been that helpful / used anyways. Is there a platform where you can post ideas and other people can build them if they think it is useful?
Starred Projects
I would like to have a collection of links (like the Github starred list) and blogposts I find interesting and so on, as a searchable database of things I found interesting and would like to (at least) be able to find again. Sometimes I remember some nice tool that I one saw but then had no use for, and I am able to find it in the “tools” list in my Github. NICE.
This is something I might actually do soonish. I am using Github Stars to keep a list of things I am interested in, I even have some lists configured there to be able to sort stuff into categories. But that can only include projects that are on Github, which is maybe 80%, hopefully decreasing. See above on why I do not use Github anymore.
I am looking into building something like this myself to host on this server. A simple list of things, searchable in clear text, with tags like #linux or #library.
Full Project Rewrites using AI
I did this multiple times over the last week, for small projects every time. Both in a work context and in private. It is not interesting enough, but I discovered a few workflow tweaks. This is one of the only things I found to work somewhat well with Claude. Having separate git worktrees, working with one claude instance on both folders at the same time. Starting an orphaned branch for the rewrite, move everything over using AI, then merge with -s ours --allow-unrelated-histories, those are really fun. But do you know what I enjoyed most? It was not the AI rewrite, it was figuring out the Git workflow that worked for me. And that is not something that I feel like writing a lot about.
AI maybe revolutionizing Software Development
Some time ago, I read this blog post about the (then) current state of AI in software development. Don’t get me wrong, the article is well written and definitely worth a read, but there is one quote that I still think about:
Until now, handcrafted software was the constraint. Expensive software engineers and their labor costs limited what companies could afford to build. Automation changes this equation by making those engineers far more productive.
I strongly disagree with this. The cost of engineers is high, but the constraint is not developer capacity. It is a communication bottleneck. Stakeholders famously struggle with knowing what they want. And that is something where AI can help, if used correctly. Build a prototype yourself, then hire a team of engineers to bring it to life.
This should, however, never mean that you can now skip iterative design cycles, user research, market research and all of the other things. I am truly hoping that, once AI is actually priced realistically, the costs will again outweigh the benefits and we can finally write code by hand like we did three years ago.
Why is this in here
I hate talking about AI in software and produt development because it makes me really angry. The amount of people that have no idea what they are talking about is a lot bigger than you think. The amount of people that do know what AI is capable of and not is probably roughly equal to zero. No AI CEO has any idea of why an LLM works the way it does, because it is just accidental math. And the fact that they all say that this will be an epic revolution is really stupid. I warned you I will get angry.
So I will probably not post about this topic that soon, unless I discover something groundbreaking that I did not see coming. I am looking forward to my “yeah, that’s what I said a year ago” moment.
CLAUDE.md vs README.md or CONTRIBUTING.md
I am not the only one thinking that we should not have CLAUDE.md with agent instructions and another file that instructs humans how to work, right? If there is a skill for claude, there could be the same things as a shell script for humans. If there is a “why did I make this design decision” in CLAUDE.md, it should also be in the README (or CONTRIBUTING / … depending on setup). Why did we invent a new file for this?
Honorary Mentions
Here is an unsorted collection of blog posts that I really like. They are not related to anything. But seeing posts like these, for different reasons, motivates me to write this here today.
- Meta Blogpost: https://dbushell.com/2026/04/01/i-quit-the-clankers-won/
- The design of this blog: https://bernsteinbear.com/blog/snprintf/
- AI is stupid: https://mariozechner.at/posts/2026-03-25-thoughts-on-slowing-the-fuck-down/