Ardour development and AI

There are no hard lines here.

ardour.org is (currently) hosted on AWS. Does that give us “the practical and ethical problems of corporate software” ?

The main post laid out a series of points about the downsides of AI that is very similar to Wiedauer’s. What I’m trying to get to is a useful, productive way to connect the downsides and the upsides. I don’t find it sufficient to say “but these downsides are so terrible, no possible upside could justify them”. It may well be that the downsides outweigh the upsides. But it also might not (just as most of us do not protest too much living in a world with lots and lots of internal combustion engines).

Perhaps it would be useful to be highly practical and specific about this. Right now, Linux distributions are steadily deprecating X Window and replacing it with Wayland. The GUI toolkit that we use has no backend for Wayland - writing one is a substantial and challenging task that almost nobody is interested in doing. For Ardour to continue being viable on Linux, then either XWayland has to continue to improve (because right now, it does not provide the same experience as “native X Window” when using Ardour) - this is not guaranteed and we have no meaningful control over whether it does or does not happen - or we need a Wayland backend.

Would that “threat” to Ardour’s continued viability justify using a coding agent/harness such as Claude to write that Wayland backend?

1 Like

AI is used everywhere, and the whole picture has changed even in the last few months. I’m not a lawyer, as they say, but I can’t imagine that the Ardour project, or anyone else, for that matter, will get into trouble for using AI help; “everyone” is already using it, and loads of them already have teams of lawyers. Anyone who might sue you or work against you for using it is probably using it too, whether they know it or not. Projects around the globe might already use, or be using, AI that includes code from Ardor. In that sense, it’s also open source, but not in the way the licenses intend.

To learn more, I have played with a “simple” form of AI over the last few weeks, or so, “Google”, which uses basically uses conventional indexing to pick up and generate code) and made a few small projects. When I ask where Google is getting the code, it always shows the sources and where it came from.

For some of the C++ tests I did for Windows, Google’s AI mode used and pointed me to learn.microsoft.com, and for some Linux stuff, Google pointed me to the Linux kernel source itself. In both cases, it also used info from cppreference.com. How it was trained to understand this info and generate code from it is another story, but aimode pointed me directly to the documents and/or places it used in order to generate or change code.

So it seems to me that, in this case, Google Gemini is using open information and documentation from the 'Net if one wants to and generate code according to one’s wishes. Is this unethical? No, I don’t think so, but the way for an AI model to get there might have been unethical, and even further. But it can already do much based on open, free, and unlicensed knowledge.

So my final advice, if I can be so brave as to say so: Use AI where it fits, and use it wisely. I will still support Ardor as I have done for many years, and will continue to do so.

2 Likes

Even from a pragmatic framework, I do not think that using LLMs will help Ardour.

One reason is because using LLMs can create a negative perception of both the resulting software and its developers. For example, the email company Tutanota had a policy that prohibited external LLM contributions. However, several people have found that LLMs were allowed to be used by Tutanota’s development team. One of the maintainers later said that they use LLMs “deliberately and responsibly”, with the output reviewed by a “human pair”.

As a result, Tutanota received backlash for their usage of LLMs irrespective of the teams “responsible response”, with multiple users stating that they have lost trust in the email service and other users stating that they will cancel their subscription (read: Tutanota will lose money) as a result of this policy. This matters to Ardour because users may also lose trust in the development team and/or cancel their subscription/donation (read: Ardour might lose money) if LLMs are allowed in development.

The other reason is because using LLMs can hinder and punish otherwise successful software. The market research platform Game Oracle looked at 9,879 paid games between January and October 2025 and compared games with disclosed AI use with games with no disclosed AI use. They found that developers that use AI see about ~53% reduction in user reviews compared to those that don’t. Games that use AI were also rated 4% lower than games without an AI disclosure.

This matters to Ardour because if it allows AI contributions, then potential users may not even attempt to use Ardour (they may hear that it uses AI from a friend or social media) and may look at other DAW’s like Bitwig 8-track, Qtractor, or Reaper; with the latter responding to negative feedback of introduced AI features by removing them. (For context, Reaper added several new stock plugins in v7.83 where the UI appeared to be generated by AI. They have received criticism for this and the plugins were removed in the next update, v7.84.)

In summary:
Even if we view LLMs from a pragmatic framework, the upsides may not offset the downsides because LLMs may harm Ardour’s reputation and long term ability to make money.

1 Like

When I was a child, I visited my grandparents with my parents. My grandfather’s (born in 1900) black-and-white TV had broken, so my dad bought him a color TV. Because of that, my grandfather never watched TV again until the day he died.

5 Likes

This reminds me of the war between digital photography and traditional light-sensitive film in the early 2000s. Back then, there were arguments on both sides - conservatives defended the traditional approach, and so on. Time has shown how that turned out. The use of LLMs in programming - or anything else, for that matter - will unfortunately follow the same path. You argue that people might stop paying for a subscription if Ardour is developed with the help of AI; however, those same people could just as easily cancel their subscriptions if they discover another tool - built using AI - that does everything Ardour does, plus much more. 90% of them won’t be held back by sentiment - much like the people who clung to film slides right up until the moment they got their hands on a decent digital SLR camera. I write this with regret, btw.

2 Likes

A similar discussion took place among painters in light of the emerging field of photography, which became widespread from 1840 onwards. They called it light painting and the representation of reality. Some photographers are considered artists today.

1 Like

Not only among them :slight_smile:

(but no, i didn’t read it, though it is mentioned a lot. I know a thing or two about the general ideas of the critical theory/The Frankfurt School. And art was quite important for them. Most of their main ideas would suite well into the discussion about llm’s in general. We mainly discuss technical and programming problems here, not philosophical ones, so this is just a sidenote - how the whole problem, technical progress and it’s impact, good and bad, on society, to me is anything but new … just like you pointed out too ).

Sure it would justify it.
Even less workload for certain tasks would, i guess.

I am not a programmer, hence can only say that much about how much workload what tasks contain. I can assure you i trust the decisions which will be made by ardour will be well justified. The existence of this very thread, see for example the opening post, seems to already show that the complexity of the problem is considered.
I hope this sounds as humble as it was meant to sound. I have no strong opionion.

Interesting talk by one of the founder of Acorn (which later became ARM, more or less). The last part is about newer technologies meant to tackle challenges of power consumption in AI. We are talking about 1/100th of the power usage, according to him.

Seems like one of the companies he mentions, Fractile, already has a contract to deliver chips to Anthropic.

If this adds up I bet we will be seeing feasible and affordable hardware for running more than adequate local AI in the foreseeable future.

Interesting (As an aside, I visited The Retro Collective where the talk took place, here in the UK a while back - really cool place with an impressive collection of vintage computing and arcade machines)

(Just for fun I also have a side-hustle coding some new 8-bit games occasisonaly - no AI involved :slight_smile: )

1 Like

Thank you for reading and responding.

Yes, using AWS is a less ethical option than most others. It’s impossible to do everything totally ethically under capitalism, but we should still try to minimize harm.

In my opinion, there is nothing that justifies voluntarily giving money to the worst people in the world, polluting poor communities, and alienating many of your users, especially when you have the choice not to.

XWayland works well enough for me, and it isn’t disappearing any time soon. It would be much better to raise funds to improve XWayland and add Wayland support to Ardour. NetBSD received a significant increase in donations this year because they prohibit LLMs, and musicians hate Gen AI more than most people.

is that “and” supposed to be “than” ?

I should say “improve XWayland or add Wayland support.”

Plug-ins will likely require X11 or XWayland or some other solution for a while too. I’m not sure how well this might or might not integrate with a native Wayland backend for Ardour? (There are proposals, and one provisionally accepted Wayland UI standard for VST3, but I imagine its going to be a difficult (and lengthy) process before anything is as commonly accepted / universal as the current X11 plug-in UIs)

1 Like

Yes, X11-using plugins will require XWayland even if Ardour has a native wayland backend. However, this is easily handled by allowing connections to the X(Wayland) server (even 1 per plugin is fine), and the challenges for XWayland posed by plugin GUIs are likely to be less than those from running Ardour itself. I mean, XWayland can basically run Ardour now, but has some edge cases that are not hard to run into.

I currently have no plans at present to adopt the Presonus/Steinberg proposal for Wayland UIs, because I think it is technically garbage and there’s a better way (all said somewhat provisionally and reserving the right to change my mind).

I don’t want to derail this thread, but just to say I think that’s a little harsh - I was assisted by one of the Presonus devs while porting some of my existing plug-ins to support the proposed standard, and who was very helpful. I think its unfair to lay the blame at the implementation of that proposal, because, at least it is a working standard (while everyone else seemed to be just hoping the problem would go away if we ignored it long enough) but also most of the (relatively few) significant issues I ran into could be traced back to some kind of limitation with Wayland itself (not least the exponentially growing matrix of compositor / implementation compatibility, and some fundamental issues with window placement and embedding built into the design or at least the design philosophy of Wayland).

I would say, as I think I have elsewhere that - it works, it’s actually not a bad fit for the existing SDK (though the VST3 SDK is another thing which gets demonized, somewhat unfairly but that’s a whole other topic), and its perhaps better characterized as the worst solution apart from all the others.

Perhaps it’s not the ideal solution from a computer science or theoretical / academic point of view, but sometimes you just have to be pragmatic and make what you need from what you can get.

That’s (probably) all I’m going to say about it here…

Requiring a host to run its own compositor? Is this real life?

Julian Wolff ? falktx and me tried to convince him at LAC 2025 to not require hosts to provide their own compositor. But he already had a proof-of-concept. I suspect corporate release schedule played some part why he moved on with it, but I don’t know for certain.

Yes. I could be wrong, but somehow I think the whole host compositor thing was kind of the “official” solution offered by the Wayland devs as the remedy to plug-in UI embedding. Not, ideal, I’ll admit (in fact I was kind of horrified when I first read the spec). But, the proposal for the VST3 standard actually includes a MIT licensed host compositor implementation. I found it interesting that there was more (grudging) acceptance once Claude could be deployed to implement a host compositor…

As I understand / remember it, I think Wayland places some restrictions on (sub)surface embedding, so (short version) unlike X11, you can’t embed a surface created on another connection to the system compositor (more correctly you can’t assign a subsurface role to a surface created on another connection), and therefore the use of a host compositor is the (only) solution you arrive at for safely forwarding the host’s connection, in a way which permits the plug-in UI subsurface embedding.

Sort of. Film simulation in digital is now very popular.