Juce A.I Agents like new Figma Make?

Development
1

I was thinking about JUCE workflow the other day. Right now most of us do our plugin UI in Figma or some other design app and then we sit down and start translating everything to JUCE manually. Takes time and you kinda repeat yourself.

Wouldn’t it be nice if there was some official tool for this… like a small visual designer for JUCE that knows all the components and layouts, and just spits out the code? Basically like Figma but JUCE‑aware. Design visually, click export, and boom you get C++ that actually runs.

Feels like it could speed up prototyping a lot and make life easier for small teams. I dunno, maybe someone tried something like this already? Or maybe the JUCE team ever thought about a design‑to‑code workflow?

Curious what people here think, I feel like this would make JUCE way more fun for modern plugin projects.

2

If you have not yet met the Melatonin Inspector, you’re in for a treat .. this is pretty close to being the tool you described - although it doesn’t generate code, it can probably be extended to do so without a lot of kicking and fuss:

3

But this is a preview tool,frankly speaking, I am referring to an AI training that is completely proficient in Juce, similar to Figma’s new Claude integration, allowing for fast designs specifically trained for Juce and writing error-free code. This would be a very practical step that would enable designs to be quickly put into practice.

4

Well, you could certainly find happy users of such a tool, were you to build it. I think Melatonin Inspector has a lot of the pieces of the puzzle needed to do it ..

5

Yes, we can actually train an artificial intelligence language model specifically for Juce. Approximately 14b models would be more than enough for this, but in order to do so, we would need thousands of working Juce codes, completed open source Juce projects, training documents, and other resources. Only Juce developers would be able to gather all of this.If trained using their own Juce source code, this would result in an amazing Juce Super Assistant and software developer.

6

Have at it - let us know when there’s something to test:

:wink:

(BTW, such use of AI/ML is not exactly warmly received around these parts, you should know..)

7

It would seem simpler to create a regular Figma plug-in that outputs JUCE source code. What benefit would AI add here?

1 Like
8

Getting VC money?

9 Likes
9

(and for anyone not tracking this, there is an MCP server for Figma, so that part is already built for you.)

2 Likes
10

I’m pretty sure once ai is making the music no one will need plugins. The only thing left we will actually need is can openers.

2 Likes
11

The world is changing very fast right now, don’t you think we should keep up?

12

3.rd party no software or plugin is ever like its own source, today when you try to write a “lookandfeel” code about Juce in Cursor, even Claude Opus 4 looks you in the face and makes mistakes all the time, Juce is very specific about this, for example, don’t you think it would be nice to have a super Juce assistant that quickly writes a midi sequencer or a deep dsp code and makes it work?

15

No, I do not think that would be nice.

7 Likes
16

Whether we like it or not, its going to happen. Kids are already out-pacing many of us with AI-built apps and software ..

1 Like
17

This sounds like you might have watched too many strange ads on YouTube. Please show us one kid with an AI-built app and software that’s actually useful.

2 Likes
18

Let me give you an example that’s very similar to this topic. I’ve been a professional DJ/Producer for 25 years. I’m 42 now, and I still play music at the top nightclubs in Istanbul and across Turkey.

Throughout my life, I’ve always heard this line from older DJs, and I’ve always known how ridiculous it is:

“If you can’t play vinyl on turntables, you’re not a real DJ!”

This is such a nonsensical mindset because it has zero meaning in the professional world. What, am I supposed to believe that if I play a record on vinyl, the sound coming out of the speakers magically becomes a different song? Or that the crowd suddenly thinks, “Ah yes, he’s on vinyl, that’s so much more authentic”?

The reality is, the audience doesn’t care about that at all. They just want to get what they came for, they pay for it, and they expect to enjoy themselves. They’re not interested in anyone’s personal fantasies or gatekeeping ideas.

This situation I’m facing now is exactly like that.

About three months ago, I started developing my own MIDI harmony engine with Cursor Pro because there was still no product on the market that truly met my professional needs. The project spiraled out of control—in the best way.

It evolved into the most powerful harmony engine in the world, with:

  • 652 scales, each one with organic code and harmonic computation,

  • Counterpoint modules, upper-structure modules, polychordal capabilities,

  • Melody-to-Harmony transformations,

…in total, 32 fully functional modules that operate with the depth of Mozart and the emotional sensitivity of Ennio Morricone. All of it is tested and working.

Now I’m at the stage where I want to integrate a GUI using JUCE. But JUCE is a very refined and conservative framework. That’s why I’m developing the workflow cleverly through Figma first.

For example, I now have a “borrowed chord brush” that can draw secondary or modal chords directly onto the sequencer. I’ve designed brushes like this, embedded them in the MIDI sequencer, and they run flawlessly under perfect harmonic logic.

When I try to bring this to life in JUCE, I run into the same kind of message again:

“AI-based suggestions like this aren’t really welcome here.”

Why?

Is it because the people at the top want to make sure the next generation doesn’t surpass them, so they can remain the “alpha predators” of the pyramid?

The world is changing in a crazy way. There are 25-year-old DJs with just one year of experience creating absolutely incredible work. And I respect that. I even guide them whenever I can to help them go further.

Right now, the best young DJs in Turkey are in my team. Even Mahmut Orhan was part of our crew 15 years ago—you’ve seen where he’s gone since.

I could have told him, “You don’t play vinyl, so you’ll never be anything.”

If I had done that, I would have been crushed by these ambitious, hardworking, and curious young people.

Instead, I choose to keep moving forward. And today, I have an amazing community and team around me.

Turns out, that’s the right path all along.

19

Let’s talk about this again in a year. Right now, I’m personally experiencing what’s possible with a platform that’s still in its infancy, and it’s already blowing my mind.

To be able to offer this commercially, I need exactly what I mentioned before—advanced JUCE agents with fine‑tuned datasets.This is just another clear example of what I mean

20

Please show us one kid with an AI-built app and software that’s actually useful.

How about a list: Adrian Martinez, InfiniLearn. Anvitha Vijay, Smartkins Animals. Kavya Kopparapu, Eyeagnosis. Tanmay Bakshi, AskTanmay. Pranav Kalyan, MathWiz. Harsha Vardhan, EcoTrack.

All teenagers … all kicking ass in their respective field by applying AI/ML skillfully.

I dare say there are some in our own realms, probably cleverly staying out of the conversation to avoid upsetting us greybeards..

21

I think it’s great that you were able to build your harmonic engine using AI (no sarcasm). If a tool like that helps you build stuff, more power to you.

The reason why we don’t want AI-based suggestions in this forum is that they are usually wrong and don’t help anyone.

Also AI isn’t always the best solution for a given problem. If you want to export Figma designs to JUCE, a regular plain-old Figma plug-in would be a better solution than an AI. Of course, you could use AI to write such a plug-in.

But it’s not like everything needs to be an AI from now on. :wink:

People on these forums and other places help each other out for free. I’m happy to spend my time helping someone to learn something. But if you’re using AI, you apparently don’t care about learning how to write code yourself, and so why should I help you? If you want me to fix mistakes made by an AI, you’re going to have to pay me for it.

10 Likes
22

There’s one thing you’re unfortunately missing.

To write a harmony engine, you need deep knowledge of harmony, and to write advanced code, you need strong coding skills. If we follow that logic, only people who already master both could ever create something like this.

I think that’s where you’re wrong.

With the way technology is evolving, so many tools are making things easier.

I’m sure that at some point, even you will use Cursor with Co‑Pilot or some other AI assistant when coding. And when that happens, if someone gives you the same kind of dismissive response you gave me—saying that your code “doesn’t really count because you had help”—it would be no different than what you’re saying now.

I used AI for advanced harmonic logic, and I personally designed exactly how the engine should function.

I don’t need to learn coding from scratch because I’m using a supportive tool.

And to clarify, I never asked you to create a JUCE AI agent for me—I simply said it would be great if such a thing existed, and I gave my own work as an example.

Don’t close yourself off to progress.

If the time comes where I need your expertise, I’ll gladly pay you.

But honestly, right now I don’t need it.