30.07.2026
I think it is no secret that I was quite happy to start to use Codeberg as my remote git repository. It was the bit of hope I had for the new age of software development that we are embarking on, now with the rise of "vibe coding", its evolution to "agentic engineering" and beyond. But something has happened recently with Codeberg that, personally for me, has removed most of the hope I had for an independent, no "shareholder value" crap, non-profit organization.
The point of contention for me is their stance on AI/LLM code hosting. From what I have read they are basically turning their backs to the current future of coding. I do understand that there is a lot of issues with LLM generated code, but the reality is that we are very much in an era where we do use the tools to write some portion of our code, repetitive tasks and boilerplate code, and to be quite frank, these are a lot of the code we do almost on a day to day basis as software developers. We are perhaps, and without throwing shade to the profession, "plumbers" that connect systems and adapts data so that it can be consumed by other services or displayed for users. Software development and architectural patterns are widespread and they are replicable, therefore they are easily instructed in a prompt if people want, therefore LLM code will exist on a lot of projects and we have to embrace it.
It seems to me that in our software developer profession we have two extreme positions to AI, and most of the people sit somewhere in the middle. On the one hand there are the developers that use it daily and use it for everything, sometimes even going at length into their prompt and basically using the same if not more time to make the same code that they are writing detailed instructions towards. On the opposing side we see the people that discard any AI as just a statistical nonsensical fad that will pass by like the "crypto wave" (no ill will intended). For most developers, and I do personally put myself somewhere in the middle of these opposing views, it is a tool that helps automate a lot of the boring parts of coding. Most of the time these boring parts cannot be easily scripted nor automated, without heavily editing the result. We have all worked in projects where there is some slight twist to a pattern. We have all certainly also encountered the famed big ball of mud projects with no structure where a script can adapt well enough the generated code. This is, to me, where LLMs shine, as they can come in and adapt the code based on the codebase and its implementation patterns.
All of this for me to say that people will use AI, of course the "no-clankers" will resist as much as possible but we already saw a variation of this happening when IntelliSense/code-completion first appeared. There are always people that are fear-mongering a bit saying that with code-completion/generation we don't learn anymore an API and read the docs, but the reality is that it's something we should do as professionals, not just because the environment forces us to do it. In the LLM age the main negative effect I see is the so called "slot-machine" aspect where you write down some words (or speak then), and then click a button to start the prompt and iterate over it using natural language, almost never going into the generated code and adapting the last bit of it to make perfect. Of course this is something that does hurt the profession, and there is also a very valid point that the effort one puts into this loop is almost null, while the effort in someone to review it will have to put in to understand the generated code and find problems with it. And yes it seems that the responsibility then falls on the reviewer, which to me will trigger some really tough conversations within teams regarding this. People stop feeling responsible for the code and they also stop learning their domain if they let themselves completely depend on LLM generative powers. We must keep ourselves honest and disciplined even though our human nature is to delegate it.
So from my own interpretation of Codeberg's new policy regarding LLM code, that you can read here, I think for my own uses I will not be welcome at Codeberg. This saddens me, and I will not say that I wish to take back any donations I have done, that it was so hastily, and not transparent at all. This voting was done in a way that the discussions that led to its outcome were not made public therefore the arguments seem to be distilled into their blog article and their change in the terms of service (reflected in the initial link of this paragraph).
I understand their reasons, specially for projects that are just also hogging up CI resources, but I do struggle a bit to comprehend the correlation between storage resources and individual projects. The reason why I don't understand this, is that I do believe most really great open-source projects aren't started with that goal in mind and are mainly just something someone did either by themselves or with a few close partners to solve their issue, and then the projects grow into something more generic and widely adopted with time. So closing the door on these small individual projects that people are building and scaffolding with AI does not make sense to me.
I do also want to say is that I wish them nothing bad, this is their decision, I do have to accept it and I have to deal with it by myself. I don't think personal attacks or even attacks to the organisation are warranted nor do I think they will change the situation. It is their decision and eventually time will tell if it was a good decision and they keep growing and garnering support or if it backfires on them.
So for my personal case, the conclusion will be that I do use, as a solo developer to start an idea, and LLM does help a lot with initial setup and start a first draft for an MVP that I can then edit and change, or even steer it to a desired outcome or close enough for me to take over. I did use Codeberg to host some of these repositories but it seems that this will not be an option, therefore I am forced to think about something else.
I am already looking at accelerating my self-hosting journey, but I am also thinking of a hybrid solution where I use an organisation that is more aligned with my values and needs such as Disroot, to which I will migrate a few of my repositories and then the rest I might try to host on my NAS and only "sync with remote" when I am at home. I was also looking into integrating Codeberg to run CI on my projects exclusively at home, but I will also accelerate such plans for Disroot too.