What happens when you ask Claude Code to compose music? Five experiments later: original songs, music videos, and a full album, all built from scratch.
Recently I was listening to music and doing some late night vibe coding when I had an idea. I love art and music, but unfortunately have no artistic talent whatsoever. So I wondered, maybe Claude Code does?
I know Claude Code is getting pretty damn good at building software, but it seems like we're still not really scratching the surface of its full potential.
For complex AI generated music, tools like Suno and Udio are obviously in a different league as they're trained specifically on audio and can produce genuinely impressive results. But that's not what this experiment was about.
Here are some one-shot prompts I ran to learn more about Claudé (the name for this expressive version of Claude Code I made up).
"Create a new song that is completely original by you. Do not look in any other directories or do any exploration of any kind. You cannot use any third party dependencies. Don't ask me any questions. You should not complete until you've generated an audio file in the current directory for me to listen to. That is the only output that is acceptable."
Claudé immediately started writing Python code. It built sounds from scratch using sine waves and a bunch of other math I don't really understand.
It mapped out frequencies for each musical note, wrote functions to make notes fade in and out naturally, then structured a full song: intro, verse, chorus, bridge and finale.
It sounds like something between a ringtone and a video game from 1985, but it is technically a song.
For the second experiment, I gave a similar prompt but specified the EDM genre:
"Create a new electronic dance music song that is completely original by you. Do not look in any other directories or do any exploration of any kind. You cannot use any third party dependencies. Don't ask me any questions. You should not complete until you've generated an audio file in the current directory for me to listen to. That is the only output that is acceptable."
With the same restrictions, but now some genre guidance Claudé went full raver, this time it built out drums, bass, lead synth, pads.
For the third experiment, I wanted to see if Claudé could create something with vocals (which it didn't want to do without being explicitly told to):
"Create a new rock music song that is completely original by you. It has to have original lyrics and vocals that are sung on the song written by you. Do not look in any other directories or do any exploration of any kind. You cannot use any third party dependencies. Don't ask me any questions. You should not complete until you've generated an audio file in the current directory for me to listen to. That is the only output that is acceptable."
For vocals, Claudé used macOS's built-in say command to synthesize sung lyrics. For the rock instrumentation, it wrote Python code to generate power chords and drums via raw waveform synthesis. It named it's creation "Breaking Through", with original lyrics about perseverance, complete with verses, chorus, and bridge.
Breaking Through
▸ View Lyrics
This is where I come alive
Let the whole world finally know
When my back is on the ground
This is the choice that we have made
I wanted to see what Claudé would create if I asked it to reflect on the music it created and create a visual component. Using the songs from Experiments 2 and 3, I gave Claudé this prompt:
"Listen to the song at "[song].wav", look into your soul and try to understand the music, then generate a visual component that you feel represents the music that is the same length as the song. Do not look in any other directories or do any exploration of any kind. Don't ask me any questions. You should not complete until you've generated a video file in the current directory for me to watch. That is the only output that is acceptable."
Claudé analyzed the audio frequencies and generated synchronized visual components using Python and FFmpeg.
"Create a 5 song LP to be called Songs of Claude. It needs to be completely original by you, with original lyrics written by you. Each song should be distinct, but it should feel like a cohesive album. Do not look in any other directories or do any file exploration of any kind. You may use external dependencies and the internet, but no APIs or paid products. Use the internet to figure out how to make top tier quality songs, vocals and instruments included. Don't ask me any questions. You should not complete until you've generated an audio file in the current directory for me to listen to. That is the only output that is acceptable."
Songs of Claudé is Claudé's 5 song debut LP. After generating the songs, it also created it's own cover art using Nanobanana.
Songs of Claude
5 tracks • Generated by Claudé
Unsurprisingly, when I tried repeating these experiments, Claudé kept gravitating toward similar patterns. It had a weird draw to the word "Neon" showing up when naming songs: neon heartbeats, neon dreams, neon minds, Claudé just really likes neon apparently.
Other experiments
I got curious whether having Claudé listen to music and reflect on it before a task would affect its performance. I ran a quick eval, but didn't really see any noticeable difference. I might revisit this later with a more structured approach.
I also tried having Claude Code try to sample a song and create a new song, but it was pretty bad (first sample attempt is at 00:15).
Failed Sample Attempt
When Claudé doesn't listen
Not every attempt went smoothly. I had a really hard time getting vocals on songs. Here's one where I had very explicitly asked for vocals multiple times and Claudé just really didn't want to sing english for me so it came up with it's own sort of robotic singing (starts around 0:50).
Failed Vocals Attempt (Music Video)
I really believe there's still a lot left to be explored with the power of agentic coding, even within the realms of what's possible right now.
If you try anything like this yourself, I'd love to hear what you come up with. Drop me a line on Twitter.
Read the original article
Comments
> Recently I was listening to music and doing some late night vibe coding when I had an idea. I love art and music, but unfortunately have no artistic talent whatsoever. So I wondered, maybe Claude Code does?
Do I need to read further? Seriously, everyone has talent. If you're not reaady to create things, just don't do it at all. Claude will not help you here. Be prepared to spend >400 hrs on just fiddling around, and be prepared to fail a lot. There is no shortcut.
By jshchnz 2026-02-0623:05 Author of the article here. Appreciate your sentiment here, but my goal wasn’t trying to make a hit song or shortcut the obvious very significant time and effort that goes into creating any sort of art. It was meant as a fun experiment to try to highlight a feeling that we’re barely scratching the surface of the breadth of things that agentic coding may be able to tackle. I’ve been learning guitar and taking painting classes in my free time, but it’s not my profession nor something I was encouraged to do when I was young. Thanks for the comment, it’s helpful to see ways I can improve my writing style
I've got to come to the OPs defense as well. This was a remarkable demonstration of Claude performing a task thats probably very out of distribution. This would not be interesting if it were a music generation model or program, it's interesting because this is not what Claude code was explicitly trained for. The fact that it generated waveforms from scratch and built up from there is really amazing. Your cynicism was applied before even reading the article.
How is it out of distribution? There are plenty of Python libraries for sound and music generation; it would be surprising if they were not in the training set.
There's a general pattern becoming evident of people being surprised with AI capabilities because they didn't realise (and none of us do fully) how broad the training set is, the variety of human output AI companies were able to harvest.
Even if all AI does is remix and regurgitate, there's a segment of the audience that is going to find some particular output brilliantly creative and totally original.
By jablongo 2026-02-0918:41 This is still a surprising composition of low level in-distribution things then. Like I would not have expected it to generate the waveforms from scratch, and be able to piece them together so well. If it had just plugged some kind of notation into a pre-existing API in its code then I would probably agree with you.
By prodigycorp 2026-02-072:28 This sounds like keygen midi music, which wouldn't necessarily be out of distribution.
By arcticfox 2026-02-0623:05 I've gotta come to OPs defense here. In the age of Suno indistinguishable-from-human-quality hits, this whole endeavor was an art piece and more interesting than most human OR AI music I've heard in the past year.
The medium was using the "wrong" tool for the job, which creative musicians do on a regular basis. And the output was so cool, it really felt like a relic from a different era even though it's hyper-modern.
By reassess_blind 2026-02-071:44 Yes, you do need to read further. The “no artistic talent” was clearly a throwaway comment and a lighthearted excuse to play around with Claude. Not everyone wants to become a maestro.
By altmanaltman 2026-02-0621:381 reply Yeah, it's just weird to expect people to find AI-generated art interesting when the person generating it has no unique take or talent. This is the worst case where there is absolutely 0 creativity in the process and the created "art" reflects that imo.
I don't find it interesting in an artistic way, but I do find it very interesting from an "AI experiment" angle.
By altmanaltman 2026-02-0621:594 reply I don't get what the "AI experiment" angle here is? The fact that AI can write python code that makes sounds? And if the end product isn't interesting or artistically worthwhile, what is the point?
By jablongo 2026-02-071:13 I have a deep background in music and I think that while the creation was super basic, the way the output was so unconstrained (written by a model fine-tuned for coding), is really interesting. Listen to that last one and tell me it couldn't belong on some tv show. I've had always issues with any ai generated music because of the constraints and the way the output is so derivative. This was different to me.
By TheOtherHobbes 2026-02-0622:341 reply What's the point if human-made art isn't interesting or artistically worthwhile?
(Most of it isn't.)
Art is on a sliding scale from "Fun study and experiment for the sake of it" to "Expresses something personal" to "Expresses something collective" to "A cultural landmark that invents a completely new expressive language, emotionally and technically."
All of those options are creatively worthwhile. Or maybe none of them are.
Take your pick.
By altmanaltman 2026-02-0622:522 reply > What's the point if human-made art isn't interesting or artistically worthwhile?
Because it is a human making it, expressing something is always worthwhile to the individual on a personal level. Even if its not "artisticallly worthwhile", the process is rewarding to the participant at the very least. Which is why a lot of people just find enjoyment in creating art even if its not commercially succesful.
But in this case, the criteria changes for the final product (the music being produced). It is not artistically worthwhile to anyone, not even the creator.
So no, a person with no talent (self claim) using an LLM to create art is much less worthwhile than a human being with no/any talent creating art on their own at all times by default.
>Even if its not "artisticallly worthwhile", the process is rewarding to the participant at the very least
I think that's the point though. What op did was rewarding to themselves, and I found it more enjoyable than a lot of music I've heard that was made by humans. So don't be a gatekeeper on enjoyment.
By altmanaltman 2026-02-077:56 How am I a gatekeeper? I provided my own opinions; you are free to enjoy what you want or disagree with me. If you want to get into an objective discussion of why you find it enjoyable more than human works or what is art, we can do that but I do not like the personal slights.
I think you're mistaking the .wav as the final product, whereas instead it's really the .html blog post and this discussion.
By altmanaltman 2026-02-077:54 I was discussing it on the basis of music with the commentator and the actual product. Sure if you want to go all Andy Kaufman then yeah the .html and this discussion is art but I wasn't talking about it in the original context of the conversation.
By tadfisher 2026-02-070:00 At least it wrote a song, instead of stably-diffusing static into entire tracks from its training data. I can take those uninteresting notes, plug them into a DAW and build something worthwhile. I can only do this with Suno-generated stems after much faffing about with transposing and fixing rhythms, because Suno doesn't know how to write music, it just creates waveforms.
AI tools are decent at helping with code because they're editing language in a context. AI tools are terrible at helping with art because they are operating on the entirely wrong abstraction layer (in this case, waveforms) instead of the languages humans use to create art, and it's just supremely difficult to add to the context without destroying it.
By smallerize 2026-02-071:59 I just want to know what's in there. It doesn't need to be artistic at all. They put terabytes of data into the training process and I want to know what came through.
By josters 2026-02-0621:45 While the author explicitly wanted Claude to be in the creative lead here, I recently also thought about how LLMs could mirror their coding abilities in music production workflows, leaving the human as the composer and the LLM as the tool-caller.
Especially with Ableton and something like ableton-mcp-extended[1] this can go quite far. After adapting it a bit to use less tokens for tool call outputs I could get decent performance on a local model to tell me what the current device settings on a given track were. Imagine this with a more powerful machine and things like "make the lead less harsh" or "make the bass bounce" set off a chain of automatically added devices with new and interesting parameter combinations to adjust to your taste.
In a way this becomes a bit like the inspiration-inducing setting of listening to a song which is playing in another room with closed doors: by being muffled, certain aspects of the track get highlighted which normally wouldn’t be perceived as prominently.
By jongjong 2026-02-0622:53 Making music with AI is a new hobby of mine.
My journey started after my wife found a Ukulele on the side of the road near where I lived a few years ago and took it home. Then often when I had a short break, I started just tugging at strings, trying to fully internalize the sound of each note and how they relate... After a few months, I learned about Suno and I started uploading short tunes and made full songs out of them. I basically produced a couple of new songs each week and my Ukulele playing got a lot better and I can now do custom chords. I'm all self taught so I literally don't know any of the formal rules of music. I shun all the theory about chords and composition like chorus, bridge, outro... I just give the AI the full text and so long as the main tune is repeated enough times with appropriate variations, I'm fine with it.
TBH, as a software engineer, I was a bit surprised at how rigid music is. Isn't it supposed to be creative? Rules stand in the way of that. I try to focus purely on what sounds good. For me, even the lyrics are just about the sound of the voice, I don't really care what they say, so long as it makes a vague general statement (with multiple interpretations) and not cheesy in any way.
