Introduction
If you have even a passing interest in music production, sound design or audio engineering, then you’ve probably come across a social media post of somebody showing off a plugin they’ve built using AI.
Previously, developing your own audio software required at least a reasonable understanding of programming, alongside a deep knowledge of music and audio theory and a well-trained ear for what sounds good. Now, in theory at least, it’s possible to create your own plugins and instruments simply by describing what you want to an AI assistant and asking it to write the code.
It’s certainly an interesting development, especially for producers who’ve got ideas for tools that don’t currently exist. But with increasing numbers of homemade plugins appearing online, I think there’s a question worth asking: just because we can make our own plugins, does that mean we should?
What Is Vibe Coding?
Vibe coding is the term used to describe using AI to write software based on prompts given in everyday language, rather than writing the code from scratch yourself.
One example would be asking an AI assistant to build a distortion plugin with a particular set of controls, or providing a schematic for a classic piece of analogue studio kit to be recreated in the virtual domain. From there, you can refine the results through further prompts, asking the AI to make changes based on your testing until you've got something resembling a finished product.
The appeal is obvious. You don't need years of programming and audio experience to start experimenting with your own software, or to recreate that classic guitar pedal you flogged years back.
However, generating some code and producing a reliable audio plugin that sounds great are two very different things.
The Potential Benefits
Theoretically, a tool that allows anybody to turn even the most avant-garde ideas into usable audio software without extensive coding knowledge can only be a good thing.
Indeed, there have already been a number of examples of enterprising AI-assisted developers using vibe coding tools to add functionality to existing hardware, including defunct Akai samplers and older models of Pioneer DJ's CDJ range.
There's also the potential to breathe new life into moribund VST plugins such as CamelPhat, much prized for saturating the master bus back in the 2000s, but discontinued following Apple's acquisition of Camel Audio in 2015.
Recent efforts to get the original plugin running on modern Apple Silicon computers
demonstrate what's possible, even if they don't involve rebuilding the software from scratch.
For me, a particularly interesting aspect of vibe coding is being able to make tools that are tailored to your specific workflow. We've all encountered situations where a plugin has every control you need for the job in hand except one, or makes a simple job unnecessarily complicated.
Beyond the practical applications, there's the creative side. Imagine coming up with an unusual idea for an effect or synth, then being able to experiment with it without having to recruit a team of crack programmers to bring your idea to life. Even if the results are basic, that process could lead to a happy accident that creates new sounds, techniques or even genres of music.
Where Things Can Go Wrong
Whilst being able to create your own software with little to no technical knowledge is an impressive feat, it creates a significant problem. Without that deep technical understanding, it can be difficult to know whether the software is actually doing what you think it should be doing.
Audio processing is, by its nature, deeply complicated. Something as seemingly trivial as a distortion plugin raises questions about oversampling, aliasing, gain staging and how the processing behaves with different programme material. Is the distortion an intended effect of the plugin itself, or a byproduct of badly written code?
AI can hallucinate, generating code that appears to be correct without necessarily implementing things properly. Having a number of close contacts in the software development industry, I'm very much aware of the effort and lengthy testing that commercial products go through before they're even available to beta testers, never mind the general public.
Reliability is another key issue. Will the plugin behave correctly at every sample rate? How about when it's used in an older version of a DAW that still supports 64-bit VST3 or Audio Unit plugins? Will it flake out when several instances are loaded, and will the parameters restore correctly when I open an old project in the future? These are exactly the sorts of issues professional developers spend considerable time testing before their software reaches our machines.
There's also a big difference between creating a piece of code that makes a sound and making a polished product that's excellent at its job. If you're creating an unusual effect plugin, a few foibles might be considered characterful, whereas if you're building mastering-grade processing, I'd argue the standards need to be much, much higher.
AI gives us all the ability to create software, but doesn't automatically give us the gift of knowing when something is wrong… or just sounds bad.
Final Thoughts
As with the democratisation of music production thanks to DAWs, virtual instruments and plugins, vibe coding has huge potential to lower the barriers to entry for software development, particularly for creating simple utilities or experimenting with unusual ideas.
Making audio software development more accessible is a positive thing, and I’ll be very interested to see what producers and engineers come up with when they’re no longer hamstrung by a lack of programming ability.
I’m also very curious to see how recent moves by manufacturers such as Akai and Behringer towards open-source firmware pan out. Being able to continually improve or alter a piece of hardware could not only lead to sonic experimentation, but also extend the lifespan of older equipment in years to come. Great for both the environment and our pockets!
But I also think there’s a danger of confusing the ability to create our own software with the ability to develop a great, reliable plugin. There’s an enormous amount of R&D that goes into developing software and firmware, and that expertise shouldn’t be underestimated.
Similarly, I’ll be watching carefully to see whether the open-source hardware movement results in a slew of new and exciting sounds, or a landfill’s worth of bricked music gear.
Personally, I’m most intrigued by vibe coding as a way of exploring new ideas, bringing defunct processors and synths back to life and solving specific problems, rather than replacing established tools that already do their jobs very well.
Ultimately, the question to ask is: does the vibe-coded software sound good, work reliably and actually improve what you’re doing in the studio? If the answer is yes, then it doesn’t really matter much how it was made.