perrygeo

That's unnecessarily dismissive. I find the statement quite enlightening and I wish other engineers would realize the wisdom here.

LLMs are pattern matching machines. Your code is the base template that will determine all patterns moving forward. No amount of context engineering or prompting can save your code from itself.

Pragmatically, it suggests we need a much greater investment in code quality, and much less hand-wringing over AI tooling.

Nix still has its share of fiddly problems and esoteric solutions. The difference is, once you (or your LLM) figures it out, you can write down in declarative code and it's solved forever. If it doesn't work, try something else or revert. Fearless system configuration.

critical point! I'd like to at least see the nature of the rejection spelled out in the policy.

Am I offending you personally by having used AI at all? Or is my contribution just running afoul of your technical judgement?

In order to get to "Is AI coding useful?" you need to personally work through "Is AI even acceptable?" - socially, environmentally, legally, biologically, militarily, technically, philosophically - I can absolutely respect those who say "hell no" to any of it!

- You can say "hell no" to contribution with any AI, due ethical concerns. Say it with your voice, not soft ban language. State your ethical objections.

- Much different to saying "hell no" to any AI contributions because of quality concerns; you know what's best for quality, security, bugs, etc.

- Much different still to saying "it depends" based on the nature of the contribution.

- Much different still to saying "yolo claude can commit to main".

They are all valid viewpoints. But deserve significantly different AI policies. Don't make me guess.

I used AI once. Therefore my contributions are forever invalid, I guess? Up to whose discretion? It would be nice to have clarity on the issue but no one seems interested in having a nuanced technical conversation. Therefore I will never contribute to your projects.

Great question. I can only assume it's the flip side: a linear accumulation of quality regressions. With no one accountable for delivering improved quality, it just doesn't happen. And while problems accumulate linearly and silently, they manifest themselves in more volatile, emergent ways like bugs and outages and outright broken software.

Yeah exactly. I'm just annoyed that AI coders hijacked my word for it. Whatever - "vibe" coding will forever be synonymous with AI coding. I don't get to decide.

I'll call my phenomenon "management driven development" or MDD - a project that runs on hierarchical control, where management decisions are made on the hour and no one takes accountability.

I used "vibe coding" pre-AI to describe companies that build software without any long term strategy or vision - aka "Agile Theater" - just pick tickets off the board and move forward in whatever random direction your manager feels like today. Your manager's vibes are the only thing that matters.

Sadly, AI vibe coding is far worse. Following your manager's vibes, there was at least some semblance of accountability. Now when an AI vibe codes something, I'm not even sure whose ideas we're vibing on, or if they belong to anyone.

To some people, success looks like the linear accumulation of small wins. Each cycle, you sprint for the goal, rinse, repeat. As long as you deliver each one on schedule, everything is fine. There is no need to create a vision or roadmap, no need to work towards or even define quality, just react swiftly to the loudest signal every day.

This is a perfect storm for enshittification. No one is accountable for the quality of what was delivered, only that it was delivered. As long as the latest fire is out, all is well. Right?

> Baby companies do just ban it outright

This might be the key point. There's legal ambiguity surrounding these licenses. Large companies with legal resources can draw their own line in the sand to protect themselves.

But small companies don't have the resources or the bandwidth to address the ambiguity. So a blanket ban (Zero AGPL and GPL for Internal or SAAS, Linux generally gets a pass) is rational.

Likewise, I see small dev shops without the resources to address anti-AI policy ambiguity are going to defer contributions until further notice.

Lol. You say no further definition is required - then proceed to outline your interpretation which is quite distinct from others. Clearly further definition is required.

If you're serious about an anti-AI policy, you owe the world a definition. It's not hard. You are making a strong statement - an outright ban. Make an equally strong case for what is and is not in scope. You must get in the weeds, since it is you yourself that is banning the weeds.

To quote Gustave Flaubert: “The art of writing is the art of discovering what you believe.” Failing to write the definition is a strong signal that no one thought through the problem.

My point is, if projects are actually serious about banning AI, write that down in enforceable language. As of today, your personal interpretation of what the policy might mean likely conflicts with other's interpretations. There's no clarity on what exactly has been banned.

Put simply, if you're so convinced that AI contributions will ruin your project, you owe the world a full and complete definition of what "AI contribution" means.

For an outright ban on "LLM, AI or agent assisted code contributions", they do very little to define what that means. This is a terribly written policy - I sincerely urge the QGIS maintainers to put enforceable guardrails on the definition.

This reminds me of the legal ambiguity behind the GPL and its definition of "derivative work". A careful analysis of the GPL shows that using GPL software doesn't trigger any other clauses. But distribution does. What is the legal definition of distribution? Who knows!?! That's why most software companies have a partial (edit) ban on GPL software. Even companies that are willing to comply with the GPL, they cannot feasibly come into compliance with such a vague definition. So it's safer not to engage.

Same will happen with anti-AI policies. Even contributors who are fully willing to do things "by hand" will start avoiding these projects. No way to know if you'll accidentally run afoul of someone's AI radar. There's nothing to protect enforcement from becoming a witch hunt.

Programming is special. When you create art, music or prose, it's an artifact that is overwhelmingly consumed directly. I enjoy the art by looking at it, listening to it, reading it. I am mostly unaffected by art I've never seen (save for maybe that which seeps into popular culture around me).

Code as an artifact is different in that it can be given agency, ie turned into software. The output of that software can be valuable despite the code being unintelligible to the vast majority of people. Very few people have read e.g. the source code for the TLS encyrption algorithms, yet billions of people's lives depend on it.