We take a cautious approach to recommending software developed extensively using generative AI, particularly new projects where reliability, code quality and long-term maintenance remain unproven.
What Is AI-Heavy Software?
We consider a project AI-heavy when generative AI has been extensively used to write or modify its source code.
Occasional AI assistance with explanations, debugging or code review does not automatically make a project AI-heavy. Software offering AI features is not necessarily AI-heavy.
Mandatory Public Disclosure
AI-heavy projects must publicly disclose their use of generative AI before we will consider them for coverage.
The disclosure must be published on the project’s website, repository README or other readily accessible project documentation. It should explain how extensively generative AI was used, what it was used for and how the resulting work was reviewed and tested.
A private declaration in a submission form or email is insufficient. We will not consider AI-heavy projects without a public disclosure.
Editorial Decisions
We assess AI-heavy projects on a case-by-case basis, taking into account the extent of AI involvement, the project’s maturity, and evidence of human oversight and testing.
We may assess established projects with proven development histories differently from new AI-heavy projects.
We reserve the right to decide which projects we cover.
Please read our Comment Policy before commenting.
Good to read.
On 23 July 2026, Codeberg changed its Terms of Use to prohibit projects that mostly consist of code written by generative AI tools, explicitly naming Claude and OpenAI Codex.
It is a shame GitHub is unlikely ever to introduce a similar rule. That would hardly fit Microslop’s determination to force AI into every part of software development.
Are you seeing a lot of vibe-coded on github?
There is shed loads of the slop.
I actually do some vibe coding. I’ve taken a useful CLI tool and removed some of the functionality I’ve not needed and added a few things that really help me. I’m not going to submit the code to the project nor make my repositirory available.. It’s just for my personal use.
fooyin has just published their AI policy:
We do not accept contributions for which generative AI was used at any stage of the development process. Please do not contribute code, documentation, tests, images, or other project material created or modified with large language models, image diffusion models, or similar tools.
Once a contribution is merged, its long-term maintenance becomes the responsibility of the maintainer and other core contributors, which is why contributors must understand and take ownership of every part of their submission.
This policy operates on trust. We may ask whether generative AI tools were used when reviewing a contribution, and pull requests that involve their use will not be accepted.
fooyin is not alone in being super strict when it comes to AI. Zig bans even talking about use of chatbot/LLM services.
Prise, a multiplexor you reviewed recently fits this, the GitHub page mentions AI assisted “collaboration” and the project is now archived.
Marco, thanks for letting me know that Prise has already been abandoned. Just to clarify, my piece was a brief overview of the software, not a review; reviews on LinuxLinks carry the Review tag.
Prise will not be included in our terminal multiplexer roundup, and has been updated to reflect its archived status.
Thanks for this! And for a really useful page!
Why not just have a separate category for AI written software?
That’s an idea for sure. But I’m still concerned about the sheer number of these projects coming out and how long they will last before the ‘developer’ gives up.
The issue is that as a project gets more complicated, LLMs can get send the individual round and round in circles without providing say a critical bug fix. The ‘developer’ themselves has no coding skills to fix it. So the likely outcome is the project gets abandoned.
Do you know that LLMs are being used for Linux kernel development?
Indeed. The kernel now has formal AI-assistance rules.
Honestly, they seem to not entirely deny AI assisted tools but rather appear to be more cautious than usual entirely human-written code. Their reasoning backing their policy is pretty sound IMO. I am myself working on tools that use AI assistance (I use it to see alternative implementations, personal preference) but I probably won’t be submitting for now, till they mature a bit more at least.
The policy gives some confidence on the listings, if anything IMO.
You’re correct. We don’t automatically veto software that makes use of AI.
I’m currently writing a review of a program (not a roundup entry) that makes heavy use of AI in its development process. One reason that convinced me to review is that the maintainer has a prominent disclosure about AI usage.
I thought about this for a while.
We don’t plan on implementing a separate category.
AI-heavy describes how software was developed, not what it does, so it is not a useful software category. There is also no reliable way to determine how much AI was used, making classification inconsistent and potentially unfair.
A separate section could give immature or quickly abandoned projects publicity and appear to endorse them.
But thanks for your suggestion, Leo.
I think using gen AI to create your own projects is perfectly fine. I’ve vibe-coded a couple of apps for my own use. Without any real programming knowledge, I’ve created something that’s really useful for me. I’m not going to share the code though.
I’ve also taken a few existing projects and added a couple of extra enhancements. They make the software so much more useful for me. Even if others might be interested, I’m not going to submit the code to either project, they are just for my use only.
I think that’s a perfectly reasonable use of generative AI.
If you can create something genuinely useful for yourself, or modify an existing program so it better suits your needs, that’s one of the more compelling uses of these tools. You know who the user is, what you need from the software, and you’re not asking anyone else to depend on it.
My concerns are much more about AI-heavy projects that are publicly released, heavily promoted and presented as mature software when the developer may not understand much of the code or be able to maintain it when something goes wrong. That’s a very different proposition from vibe-coding something for your own use.
I really admire projects that completely ban AI.
This article has been adapted into a formal policy.
I don’t give a monkey’s whether software is written with AI.
I agree. In the same way I don’t really care about the language used to write the software, what IDE the developer uses, and so on.
I notice that Flathub is now following your policy by allowing AI coded apps where clear disclosure is given to users.