Human stewardship matters more than who typed the code [LWN.net]

Human stewardship matters more than who typed the code

Posted Jul 24, 2026 11:17 UTC (Fri) by grothesque (guest, #130832)
Parent article: Codeberg: Protecting our FLOSS commons from LLMs

I am disappointed by Codeberg’s decision and find it harmful and quixotic. I believe that they discount of projects with “development teams of none” too easily. There exist many useful but niche projects that are starved for manpower. And there exist many people with ideas and experience, but without large teams.

Here is one concrete example: Together with colleagues, I have created a library (https://pypi.org/project/kwant/) that has become pretty successful in its particular domain of quantum physics: since its initial release in 2013 “Kwant” has been used in the preparation of at least 1000 scientific publications by physicists all over the world. Even so, we never managed to create a self-sustaining open-source community around this project. This is not surprising: even widely used general-purpose libraries struggle to attract contributors. Our library has a narrow field of application, and advancing it requires expert knowledge that only a few dozen (very busy) people have.

Kwant itself does not fall under Codeberg’s restrictions, but it shows that useful software does not necessarily have a large team behind it. LLM-based agents are a welcome help in developing and maintaining such niche projects.

As a researcher, I only have a small team of people. Understandably, these young colleagues are mostly focused on their research and not on code quality. I, however, care greatly about code quality. And I am confident to have the necessary experience, having started as a teenage coder in the 1990s, and having continued this journey ever since. (There was a time when I took pride in hand-crafting x86 assembly. It was fun, and I learned a lot, but this is not an efficient way of writing all code. Why do people believe that one must hand-type all the code that one cares about?)

*Reasonable* LLM use is empowering. To me this means using Pi (https://pi.dev/) as a sparring partner for developing and challenging new ideas. (I value human sparring partners, but these are not always available, especially for tedious tasks. LLM agents are also useful in their own way, for example by having extremely broad, if fallible, encyclopedic knowledge.) I use agents for accelerating code review, debugging, and refactoring.

Here is a non-scientific project that I was able to realize recently with LLM assistance: a frontend to Bubblewrap that streamlines the creation of ad-hoc sandboxes directly on the command line (https://github.com/grothesque/sandkasten). This project is still young, and I haven’t advertised it widely, but I believe that it has promise. In a narrow sense, most of its code lines have been generated by an LLM, but I have nevertheless spent more than a hundred hours of my human time on this project. A lot of human thought went into every aspect of it. I would not have had the time to create something like this manually.

I considered moving this project over to Codeberg. Not anymore.

Mindless vibe-coding is not the only way of coding with LLMs. I do see the problems that Codeberg mentions, but forbidding LLM-generated code is not the right solution in my opinion. Codeberg’s terms of use classify projects based on how their code lines were typed. I believe that what matters is genuine care and good engineering.

to post comments

Human stewardship matters more than who typed the code

Posted Jul 25, 2026 7:05 UTC (Sat) by emk (subscriber, #1128) [Link] (7 responses)

Those sound like some really cool projects! And mad respect for being one of a few lonely maintainers of a piece of scientific infrastructure.

I, too, was puzzled by Codeberg's seemingly weird attitude towards solo developers that kept cropping up in phrases like, "It seems like many ‘vibe coders’ don't realize that they don't actually have a community around them." I mean, there's a surprising amount of code in major Linux distributions developed by people who don't have "a community around them." The brutal fact is that XKCD is right (https://xkcd.com/2347/), and a lot of foundational code is kept running by one unpaid volunteer. Or it's effectively abandoned, and the distros pass around patchsets to keep it building, with no upstream to ever merge patches. Community can be a damn scarce thing in the open source world sometimes. Goodness knows that 1,000 stars on GitHub frequently results in 3 drive-by PRs in 5 years.

Personally, while I have real concerns about where this whole AI thing is going, I'm genuinely happy to see end users using LLMs to throw together personal software. Now they have software that does exactly what they need, that doesn't spy on them, and that they can customize. When I got involved in free software several decades ago, that was originally the dream: Users in control of their own systems, with the freedom to change things! But now there's apparently a political movement in the "open source" community which is aggressively criticizing these users for figuring out a way to put together custom tools.

Sure, frontier models are way too electricity hungry. But we've reached the point where there are MIT-licensed, open-weight coding models that run on 5 year old gaming GPUs, and that use about 280W when generating. They sure aren't Fable, but they do actually work. Stick 'em in a read-only sandbox (like your tool supports) and they can even help troubleshoot a Linux system, keeping all the data 100% local. It was a revelation the first time I asked a local model, "Ugh, what's wrong with systemd?", and it actually figured it all out.

I do dread many of the places this all might go. But seeing users take control over their computers? That's the one undeniably bright spot, to my eyes.

Human stewardship matters more than who typed the code

Posted Jul 25, 2026 13:02 UTC (Sat) by kleptog (subscriber, #1183) [Link]

> Personally, while I have real concerns about where this whole AI thing is going, I'm genuinely happy to see end users using LLMs to throw together personal software. Now they have software that does exactly what they need, that doesn't spy on them, and that they can customize.

This. I have colleagues who are technically inclined but can't really code, using an LLM to write small tools to make their lives easier. In one case a simple bottle server that showed the availability of the two charging points near their house. The LLM had found the API specification and made a simple web application with auto reloading, version control and other features which I wouldn't have gotten around to because I'd have gotten bored before that point. It helps them debug their HomeAutomation setup.

These are the real beneficiaries of LLMs. The creatives who had the ideas but not the skills to make it happen. I guess we're feeling like the master craftsmen felt when interchangeable parts, machine tools, and later CNC machining lowered the barrier to making useful things. Look at what 3D printers have enabled.

Human stewardship matters more than who typed the code

Posted Jul 25, 2026 20:10 UTC (Sat) by Cyberax (✭ supporter ✭, #52523) [Link] (1 responses)

There's nothing wrong with solo projects that exist to scratch an itch.

It's just that Codeberg is not a place for them. Each project adds a bit of administrative overhead, and they don't have resources to maintain infrastructure on the scale of Github. And this has been the case from the very beginning.

I'm a bit amazed by this storm in a teacup, TBH.

Human stewardship matters more than who typed the code

Posted Jul 25, 2026 20:21 UTC (Sat) by mb (subscriber, #50428) [Link]

> Each project adds a bit of administrative overhead, and they don't have resources to maintain
> infrastructure on the scale of Github. And this has been the case from the very beginning.

Yeah. Nice straw man.
This is not at all what this is all about.
Here's what they claim this is about instead:

>Codeberg: Protecting our FLOSS commons from LLMs

With a very weird definition of "protecting"

If they don't have the resources, it's fine. They can ask for payments or ask me to leave or whatever.
But instead they ban everybody using LLMs. Totally different thing.

Human stewardship matters more than who typed the code

Posted Jul 25, 2026 21:01 UTC (Sat) by rbranco (subscriber, #129813) [Link] (3 responses)

> I'm genuinely happy to see end users using LLMs to throw together personal software. Now they have software that does exactly what they need, that doesn't spy on them, and that they can customize.

None of that matters to the upper-middle-class, college-educated Xennials from rich countries in Western Europe & North America. They're privileged enough to externalize their morals onto others. The ban on cryptocurrency projects shows they're way out of touch with the reality in other countries where this is the only way to earn money and to avoid hyperinflation. They're a tiny but noisy minority.

> When I got involved in free software several decades ago, that was originally the dream: Users in control of their own systems, with the freedom to change things! But now there's apparently a political movement in the "open source" community which is aggressively criticizing these users for figuring out a way to put together custom tools.

The good fight will be about having open models and bringing down hardware prices! Everything else is noise and burying one's head in the sand. But this fight can only be won by pragmatic people who are not afraid of getting their hands dirty, not a hipster commune of unpolluted monks.

Human stewardship matters more than who typed the code

Posted Jul 25, 2026 21:06 UTC (Sat) by Cyberax (✭ supporter ✭, #52523) [Link] (2 responses)

> The ban on cryptocurrency projects shows they're way out of touch with the reality in other countries where this is the only way to earn money and to avoid hyperinflation.

Fuck cryptobros. You're nothing but a scourge on humanity. Nothing good came out of you that is even in the same order of magnitude as the harms caused by cryptocrap. That include even large-scale human trafficking for the "pig butchering" scam farms.

And don't you dare pretend to be sanctimonious moralists. You do not provide a unique protection against hyperinflation anywhere.

Stop it

Posted Jul 25, 2026 21:10 UTC (Sat) by corbet (editor, #1) [Link]

OK, all of you, stop this. Nobody gains anything by you all throwing mudballs at each other. There are plenty of sites where that kind of stuff is welcome, but not here.

Human stewardship matters more than who typed the code

Posted Jul 25, 2026 22:02 UTC (Sat) by rbranco (subscriber, #129813) [Link]

> Fuck cryptobros. You're nothing but a scourge on humanity. Nothing good came out of you that is even in the same order of magnitude as the harms caused by cryptocrap.

I'm not a "cryptobro" (whatever that means). The correct term is Cypherpunk and I'm not one.

> That include even large-scale human trafficking for the "pig butchering" scam farms.

Every technology can be used for good and evil.

> You do not provide a unique protection against hyperinflation anywhere.

Tell that to people that use Binance to send money to their families overseas.