Plugin pricing in the AI era

[Tiles]

I don't disagree with the USCO statement itself. I disagree with how broadly you apply it.

The key phrase is "the mere act of writing prompts." My point is that software development is fundamentally different from prompting an image generator. A complex software prompt often contains specifications, algorithms, interfaces, architecture, error handling and behavioural requirements. At some point, the distinction between "mere prompting" and software design becomes blurred.

There is also a very practical aspect. AI-assisted coding is now used throughout the software industry. GitHub Copilot, Cursor, Claude Code, ChatGPT and similar tools have become part of everyday development workflows, from individual developers to large software companies. If your interpretation were correct, it would call into question the copyright status of an enormous amount of modern commercial software.

That alone suggests the issue is far from settled. The question of where "mere prompting" ends and human authorship begins in software development has not yet been clearly answered, especially for complex software projects.

[pekbro]

Then they should copyright the prompts. To me the copyright policy seems to have validity, in many cases.

[antto]

soon: Our Product is made with Patent-pending Prompts (wow! that's good! right?)

[pekbro]

Well the whole situation is stupid, before long there will only be a need for code wranglers, even that could be done with AI.

[stoopicus]

So in my opinion, there is no such clear category as "purely vibe-coded plugins" in the same sense as AI-generated images. The boundary between instruction and authorship is much less clear in software, and that is exactly the point that still needs clarification: it is not settled yet.

yet the USCO has a clarified position:

Prompting is Not Authorship: The Office has explicitly stated that the mere act of writing prompts (like those used in Midjourney or ChatGPT) does not make the user an "author" of the resulting AI output, as the prompt is simply giving instructions to the machine.

..but yes until its tested in some way we can all hold to our opinions.

Given that over the course of the last year, nearly all mid to large sized software companies (i.e. the ones with lawyers and money) have moved to heavily incorporating AI into their workflows, I wouldn't hold my breath for any great liberation of the finished products.

[ghettosynth]

So in my opinion, there is no such clear category as "purely vibe-coded plugins" in the same sense as AI-generated images. The boundary between instruction and authorship is much less clear in software, and that is exactly the point that still needs clarification: it is not settled yet.

yet the USCO has a clarified position:

Prompting is Not Authorship: The Office has explicitly stated that the mere act of writing prompts (like those used in Midjourney or ChatGPT) does not make the user an "author" of the resulting AI output, as the prompt is simply giving instructions to the machine.

..but yes until its tested in some way we can all hold to our opinions.

Given that over the course of the last year, nearly all mid to large sized software companies (i.e. the ones with lawyers and money) have moved to heavily incorporating AI into their workflows, I wouldn't hold my breath for any great liberation of the finished products.

Exactly so.

[TechHaus]

2029 - hoping Lina Khan is back at the FTC and we have a whole cadre of advocates in government to fight for the consumers, workers, and copyright holders that were wronged.

[gearwatcher]

I think you are presenting a straw-man. I said nothing, in my original post about code that was substantially authored, or modified by a human, I was ONLY talking about clearly VIBE-CODED plugins, which is what the poster I was responding to was referring to...

Even that is very questionable tbh.

Are you essentially suggesting that legal departments of software shops among the Fortune 500 companies, you know, the kind that even push people to use AI assisted code generation more than they personally would with dumb metrics and reward/punishment corporate bullshit, are too stupid to notice that glaring copyright issue?

I can assure you, having been part of such processes irrespective of AI (and way before this LLM explosion of recent years), that there is very little other than software IP rules across the globe that keeps these people (legals in software biz) awake at night, so I'd judge that this being any kind of an issue at all is about as probable as me winning Euro Jackpot (I never played it, for full context).

I have never seen a single shred of inclination that AI generated software development fails any prerequisite to be copyrightible, and the widespread practice of software shops left and right push for more AI codegen, AI code reviews, AI bug hunts. Claude Code may not be the most valuable employee, but is certainly the most industrious one, in pretty much every software enterprise.

My feeling (IANAL) is that this ruling is taken to mean copyrightable artwork in the narrower sense, i.e. music, images, literary art etc., and simply hasn't been interpreted by anyone in any significant capacity to extend to software work of any kind.

There are shitloads of problems with LLM generated code (code quality, fitness to purpose, security issues, hidden bugs) and these are being talked about and adressed industry-wide. There is even an IP problem of having to recognize accidental replays of someone's copyrighted code, and this is a real IP issue, one that does get mentioned and it worries people because it does open one to litigation. But this is the first time I'm hearing about copyrightability of LLM generated code being mentioned as an issue on the grounds of that ruling.

[stoopicus]

And then there's the question of even detecting that something was coded using AI vs entirely by hand. Good luck with that for highly optimized code like audio.

[pekbro]

That shouldn’t be too hard actually. Because of the fundamental way it works. Also the fact that computers really can’t do the random thing. Only pseudo random.

[stoopicus]

That shouldn’t be too hard actually. Because of the fundamental way it works. Also the fact that computers really can’t do the random thing. Only pseudo random.

Please describe in detail how you think this will work for compiled, optimized object code from AI-generated C++ - because I guarantee it won't be as easy as you think

[pekbro]

You would have to have to do some reverse engineering at times of course. Anyway, I don’t think it’s as impossible as you think. But whatever.

[stoopicus]

So you were just guessing then? Fair enough, but to be honest the current models are generally doing a great job of generating code that is pretty indistinguishable from humans; it's also not like they are doing large scale copy and paste (which IS detectable, but is also something humans do when plagiarizing, and AI models generally do not.)

[stoopicus]

tl;dr, for coding they are getting scarily good, it's the more senior developer tasks like design and architecture where they shit the bed

[pekbro]

If the gov is doing it, then they just have to force them to provide the code as well.