That sure is a lot of plug-ins on linuxdaw.org that I have never heard of. I was wondering regarding the art craft programs, I’ve seen people commenting that you shouldn’t use them because they’re not safe. Why would they not be safe if they are open source and the community can check the code. I would imagine that they could be buggy, but they are open source.
Just because the code can be checked doesn’t mean it has been.
2 Likes
Check back in a year and see how many non-vibe-coded plugins are still being maintained and supported. This problem isn’t unique to vibe-coded plugins.
Just because the code can be checked doesn’t mean it has been.
Again, not unique to vibe-coded plugins. The same can be said for any open-source software, which most users do not have the technical ability to check, even if the code is available to them.
… and that’s the reason, why I never would use SW without a developer or a team of developers behind for any more or less serious work.
Support!
5 Likes
True, but there are additional, enshitifactory, business pressures associated with the gen AI tools, so it might be cost effective for individual developers to make (free) stuff with them while we’re in the early stages of the familiar cycle, but when we reach stage 3 and they start extracting shareholder value, the costs may become less attractive. That’s not necessarily the case for a maintainer / developer of a conventional open-source project.
Unfortunately the two are kind of correlated, in as much as the rise in AI generated software may disincentivise those developers (or force commercial companies out of the market) and therefore even non-AI generated projects may also disappear precisely as a result of AI usage.
AI can be a significant enabler, and has been, of many important scientific advances. I’m still not comfortable with the socioeconomic and environmental costs, but you - could - argue a scenario where that might be a price worth paying. But generative AI has no good side.
More positively, the relatively steep increase in (all kinds of) plug-ins for Linux should be a good thing - its where I hoped it could get to when I first started using Linux (coming from Windows) for audio production in the early 2000’s. If a few of them are vibe-coded, and / or a bit rubbish, it might not matter - its still building momentum, and eventually the higher quality and better supported products will rise to the top.
2 Likes
Coding used to be hard, time consuming and expensive. Now its easy, fast and cheap. This will lead to an explosion of open source software. Closed source is not protected anymore by the cost of development. Small startups and individuals can now create comparable products fast and very cheap. I think open source will be the winner here, since it may not be possible to copyright or own AI created code.
The new workflow for a complicated project is now: have a back and forth conversation with the AI to create a complete specification of what the code needs to do. This is now the most time consuming part of coding. Then you let AI to create the code and tests for it and after that you review the code with one or two different AI’s to check for bugs, security and if the code follows the specification. Then make AI fix any bugs or deviations from the specs and then you test things manually. Then rinse and repeat to add functionality one by one.
I will continue to code by hand because I enjoy the challenge and the new way of doing things won’t take that away from me. I can choose when and how to use AI and when not. One can for example use it instead of reading stack overflow to explain some specific concept / problem I face when coding.
Not necessarily. Claude is expensive. I know people here in the Berlin audio community who easily spend 1k€ on it every month. And unless you know what you want and have at least a basic understanding of Sound, DSP and ears to check, the result is crap.
4 Likes
Claude is not the only option, you can use local AI (Qwen 3.6 / 3.8) or the open source AI model “Mimo”, that is dirt chip at the moment and very very capable and on par with the frontier models.
Model capability tests versus cost measurements can be seen on this site. Everything in the green rectangle is cheap and capable. Claude is very overpriced compared to its capabilites.
You still need $$$$ of computer hardware to run it locally, and the electricity bill for that hardware which can be significant depending on how heavily you drive it.
Cheers,
Keith
1 Like
I’m really curious as to how much utility I could get out of a local model. I don’t have the hardware to run the bigger models, and the smaller ones that I have tried locally are disappointing. I do like the idea of being able to use models in private.
I wish I were as confident as mike-acmtaudio about where this is all going. As a software developer part way through a career, I keep waffling between two thoughts:
- My skill set is being devalued and I’m going to have to figure out a career pivot in the next 5 years
- My skill set will be more valuable than ever because either the AI bubble will burst or Jevon’s paradox will reward those of us that understand how computational systems actually work.
Meanwhile I’m working on my new collection of audio plugins mostly writing the code myself just because I want to.
1 Like
I suggest to know your enemy 
At the moment it makes more sense to use some of the cheap + good cloud models than to buy a GPU. For the price of a good GPU (RTX 3090) you can use these cloud models for many years. And with that GPU you will still be limited using the small less capable models. I know. I bought that GPU just to learn about AI since I need that knowledge in my IT job.
Start by testing how coding works with Mimo or GPT 6 Luna (both are very cheap but good).
The key to understand here is that AI can’t create anything useful from one prompt. There are too many ways to interpret words and if there is just a little room for misinterpretation AI will probably take that wrong turn and go down the rabbit hole. One needs to specify very clearly what the code must do and think ahead of corner cases and cover those in the spec also. And it is a process of code, review, refine, code, review, refine. It takes time and effort to get to the goal. There is no free lunch here either.
One maybe a bit boring but very informative video describing this process is here : https://www.youtube.com/watch?v=1Fl94dPn2mU
From this video you get the feel of what it is about when you use AI for real development and not just vibe coding.
New and more capable models come out every week so I decided against getting a subscription from one specific one and be tied to that. Instead I use OpenRouter that is a service that is a gateway to a lot of cloud model providers and you can choose which one you want to use your credits for today. There are other services like that also.
For quick tests you can get a free web browser - only account from most of the providers (Anthropic, GTP, Minimax, Qwen, Gemini, Kimi, etc. Just create an account and you can use those services to test them out. There are usage limits, but I never hit them.
1 Like
What about a developer (or team of developers) that uses AI as a force multiplier? What are your thoughts on that?
Historically, the major impediment for open-source audio development has been manpower. It’s the reason why Ardour (as awesome as it is) isn’t nearly as feature-complete as something like Logic or even Reaper. There’s only so much Paul, Robin, and the other contributors can do without the kind of resources that a company like Apple or Cockos has.
AI excites me because it enables one person, or a small team, to do the work of a much larger team. While there are definitely valid concerns about it, I remain optimistic that it will help to fill in the gaps that currently exist in the open-source audio world.
Earlier this year, we got PitchNet, which is the first serious FOSS Melodyne alternative. Right now the Surge Synth Team is using coding agents to revive SpectrumWorx, a former commercial spectral plugin that was discontinued in 2016. And Marc from Dusk Audio is continuing to make high-quality plugins with the use of coding agents.
As long as the person using the agents knows what they’re doing, I don’t care that they’re using them. What I do care about is whether the software does what I want it to do. And thanks in part to the use of coding agents, I’ve gotten a lot of great open-source plugins this year that have enabled me to do things I couldn’t do without them. I want that to continue.
Note: Cockos has the same personnel resources as Ardour. We just suck (more).
5 Likes
Robin and my objections to AI coding agents are not that they don’t work (though they work less well on Ardour’s code than some might think). The objections are to the political, social, economic and environmental impacts of LLMs as currently being developed and deployed by a variety of companies and organizations.
If you don’t care about those impacts, or think that they are overblown, then sure, using coding agents to augment development makes a lot of sense.
9 Likes
Like I said, I think these are valid concerns. But as a user, I also get frustrated when you can’t do something in Ardour that you can do in other DAW’s. If someone vibe-coded an open-source, Linux-native Logic clone with all its features I’d definitely try it out just to see if the experience is easier or more enjoyable.
Note: Cockos has the same personnel resources as Ardour. We just suck (more).
You guys don’t suck. Ardour’s an amazing project overall, it’s just lacking in some areas. I appreciate all the work you guys do, even if I’m not completely satisfied with the program.
1 Like
We have a list of at least 50 things you can do in Ardour but can’t do in other DAWs. Does that count?
You should try out faderframe.
1 Like
I agree with you. There will be consequences as this is a disruptive technology. I am a knowledge worker in IT and AI will impact this sector in many ways.
There has been similar changes before for example in the beginning of the industrial age when people moved from rural areas to cities and started working in factories. New machines made agriculture more efficient and ended many jobs there. But at the same time the same industrialization created new jobs in the cites.
Then computers were invented and removed many jobs but at the same time the need for people with computer skills emerged: punch card writers, operators, programmers, repair specialist, typists, etc.
It think something similar is happening now. There will be new jobs for those that can understand AI and bend it to their will. Many jobs wont go away but will just absorb AI as an assistant and make people more productive.
For the environmental impacts I think AI as we have it today is a resource hog. This largely because we don’t understand exactly how AI works, people just stumbled upon this new technology when trying to emulate neural cells. What we have now is a rough first version of a technology that works but uses too much resources. As time goes by it will be optimized for resource usage because competition will drive prices down and AI providers will try to shed every expense they can.
We used to have very resource and energy hungry mainframe computers in the beginning. But then decades of optimization has led to low power computers that are many times more powerful. The same will happen with AI, I believe.
I tried out the Windows version (I don’t have a compatible Linux machine available at the moment) - looks very impressive, and it seems to work, but I crashed it within a minute of using it (added another track, moved the audio on the timeline a bit, I was about to try an edit and it quit, I don’t know why. Also has random GUI freezes and hangups when opening (built-in) plug-ins). Not quite ready yet I think - and it also asked for permissions to access my network for some reason (I denied it).
The Luddites were a (later much maligned) group who understood that new technology (in their case, looms) would bring about greater productivity and efficiency, but also understood that the way that technology was going to be deployed would be damaging to large numbers of people, possibly across generations. They refused to participate in the deployment or use of that new technology for this reason (and in some cases, actively destroyed instances of it).
The fact that in another 50 years, LLM technology (or some broader category of the same general concept) may use more appropriate amounts of energy and may help some of the same types of people that it will harm today does not, for me, require me to acquiesce and participate in its deployment and use.
The technological revolutions of the past did eventually bring some kinds of benefits to large numbers of people, but not before many of them, across generations, had been hurt both physically, psychically and emotionally, not before it destroyed communities, and not before or after it required rapacious consumption of natural resources to make it real. Are we better off now that we have industrial agriculture, automobiles, refridgeration, highly automated mass production, telecommunications, computing infrastructures etc. etc ? There is no need to provide an answer to that to still acknowledge that whether the answer is “yes”, “no” or “maybe”, the costs along the way have been considerable.
9 Likes
Yes I’ve heard about the Luddities 
They did face a change like this and I don’t envy them.
I agree, there is no way of foreseeing what this change will bring with it. Some of it may be bad, but I just chose to look at the bright sides of it. I may be completely wrong here, but I hope the good things will outweigh the bad ones in the long run.
If all goes well one thing that might be beneficial to people in developing countries is that AI might be used as a teacher where there are no resources to hire one. This technology may bring education and knowledge and hopefully lift people out of poverty. But who knows only time will tell how it goes. The technology holds a big promise though.
But at the moment it is “best to know the enemy” and get a understanding of what it is all about and what it can do.
If someone is interested there are a two youtube channels that report what’s happening in this field. The pace is incredible and mainstream media is mostly not documenting it (where I live).
This one documents weekly every major advancement: https://www.youtube.com/@theAIsearch/videos
This one looks at things more from the human perspective: https://www.youtube.com/@AIForHumansShow/videos
If you have not followed what has happened in the last 6 months prepare to be amazed of where the technology is now.