Made with AI. Here's what that means.
I'm going to say this plainly because the alternative is to leave it unsaid, and that is not how I, or this company, work.
The three free synths, Mellt, Tant and Tonnau, were written with the help of AI coding tools. So was most of this website. I designed the instruments: what they are, what every control does, every range and default, the pin matrix, the algorithms, the Era knob, the look. I specified, I tested, I threw away what didn't work, and I did that for about 300 hours by release day, give or take.
The code was written with an AI assistant working from my specifications, and checked against them: the manuals are written from the code and the code is checked against the manuals. But more than that, the concept, the core idea to pair vintage interfaces with digital sound engines, was mine, and without it no amount of prompting an agent would have resulted in the instruments we are discussing here.
What I think about it
What nobody needs is more AI slop. But we also need to be clear that whoever you get your software from, from now on, some of its code will have been written with help from AI tools. In fact that doesn't just apply to software: the car you drive, the fridge that cools your beer and the phone in your pocket will all have AI input at some point in their design and production process.
That being the case, a knee-jerk rejection of all things AI-related simply limits your choices to what we will soon think of as vintage code, produced by humans alone, with all the frailty, bugs and lack of vision which often came with it.
Those of us still interested in sorting baby from bathwater will want to continue using the latest updates to our familiar tools, and will stay open to new tools when they arrive.
The music industry has many very legitimate concerns about AI. Especially for artists, there are red lines to be drawn.
What we need are clear quality standards. The deciding factor stops being who wrote the code and becomes how good the code is at delivering on the manufacturer's promises. Does it add value in the context of my existing workflow, and/or help me improve my current operation overall? Does it help me get musically useful results I couldn't get another way?
Why we don't do analog modelling
This is one of the many reasons we don't do analog modelling. Rebuilding what we already have in the analog domain simply doesn't give us a big enough lift. For sure there are some small conveniences: presets and easy recall, speed to test an idea, no cables that need soldering. But the analog version usually sounds better.
Our focus has been on bringing out things that are best done digitally. We have consciously unpicked the analog concepts we have inherited from our industry's history and replaced them with their digital equivalents.
On its own, this allowed us to push beyond outdated modes of thinking. We've thought about modulation, FX, filters and voice architecture in a new way, designed to squeeze every drop of functionality out of our code without putting too big a load on your CPU.
AI beyond the code
In addition, we've tried to use AI tools outside the code-building process, in ways that add value for a small software house with limited resources. I don't have time to read every new paper written in every maths journal looking for new ways to create oscillators and/or process audio. I don't have time to distil the learnings from those readings into testable hypotheses which we can experiment with. I don't have time, but the process adds real value, and now I don't need time.
This is a good example of what we see as good practice: use AI tools to do the huge, complex, data-gathering tasks and play back the learning. Keep people in the creative roles, applying that learning to new ideas.
What it made possible
I built three complete synthesizers and I can give them away, because there isn't a staff bill to carry in the meantime. Development cycles that would traditionally have involved a whole team were compressed and focused on my vision alone. What we lose in collective endeavour we gain in clarity of focus. This technology opens instrument-making to people who know exactly what they want and couldn't previously afford to build it. I think that's good for the community in the long run, and I'm constantly surprised by those who would normally back the little guy siding with the big development houses on this issue. If you're anti Big Tech but opposed in principle to AI, I'm not sure you have yet reached an internally consistent position. Open-source AI may be one of the most democratising forces we have ever seen.
Whatever the future holds for Silicon Antiques, I've built three instruments I'm proud of, which sound the way I wanted them to and bend further than most synths I know, and which are not only free but will get regular updates and bug fixes for at least two years. That is a new paradigm. And it brings real value to our community. None of us want to see skilled people put out of work. That risk is real and I'm not ducking it. But if we also see an explosion of boutique, sole-trader designers who can now realise their vision for a budget that doesn't require remortgaging the family home, then there may well still be significant net gains for artists who rely on new instruments to keep things fresh and engaging.
What "regular updates and bug fixes" means in practice, so you can hold me to it: if you report a bug that crashes a synth or loses your work, and I can reproduce it, a fix goes out within thirty days. Whatever else happens, each synth gets a maintenance release at least every six months, with a Workbench entry saying what changed. That runs from the day each synth was released, for at least two years.
Where the free synths fall short
The panels are functional rather than beautiful; a designer would do better, and, one day, one will. And at release they are Mac only and Apple Silicon only; a universal build follows, Windows is further off.
Where they don't
I'd put the sound engines and the modulation up against synths from established makers, and I'd rather you tested that than took my word for it. In the end the rubber meets the road in the studio or on the stage. If these instruments don't give you something you can't get elsewhere, they have failed and so have I. But if you use them on a tune or on stage because the interface has educational value for you, or because you can get a sound you want quickly and for free, then, in my view, we've both had a return on the investments I have made. The download is free; the specs are on each page; the Workbench log shows what shipped and what slipped.
Quality first
We will all want to avoid software generated via short, imprecise and poorly structured prompts, in the same way we already avoid websites we suspect have been the result of "make me a website that does that" style inflections.
But most of us will want to embrace a world where people with great ideas are empowered to realise those ideas through rigorous testing, careful design and disciplined execution, even if AI tools were involved.
Quality and efficacy remain the key deciding factors.
Andy White, 20 October 2026, updated 29 October 2026. [email protected] if you want the debate.