Codex is an underrated music production tool. I've been using it to build my first song, and I'm surprised OpenAI hasn't showcased this use case.

I've been experimenting with Codex for something I rarely see discussed here: actual music composition and production. I recently made my first song using a combination of ChatGPT, Codex, Suno (for vocals), and open-source audio tools. What started as an AI-generated music experiment has gradually evolved into something much more interesting: a workflow where we're taking increasing control of the actual composition and instrumentation. One example is the piano arrangement. I developed it separately from Suno, using Salamander Grand Piano, a free sampled Yamaha grand piano instrument. That got me thinking: Why couldn't we use Codex to build entire instrumental compositions from scratch? The workflow I'm exploring is: ChatGPT as the composer: Develop melodies, chord progressions, harmonies, and sheet music using structured notation such as MusicXML. Codex as the implementation engine: Translate the score into MIDI, handle note timing, velocity, sustain pedals, and expressive performance details. Virtual instruments: Render the MIDI using sampled instruments like Salamander Grand Piano, without needing Suno to generate the audio. GitHub as the composition workspace: Version-control the scores, MIDI, rendering scripts, and revisions. Iterative refinement: Have Codex change a particular passage, introduce counterpoint, adjust voicings, or modify the performance without regenerating the entire song. Imagine prompting Codex: "Rewrite measures 33–40 with a more emotional right-hand countermelody, preserve the existing left-hand harmony, adjust the velocities for a more natural performance, and render three variations for comparison." This isn't fundamentally different from asking Codex to modify a specific function while keeping the rest of a codebase intact. The difference is that the output is music. And because the composition is represented as structured data and code, the entire process can be reproducible, inspectable, and version-controlled. I'm not claiming this replaces a DAW or a professional musician. Musical judgment, expressive interpretation, and critical listening still matter enormously. But the potential for AI-assisted composition with precise control over individual instruments is substantial. What surprises me is that Codex is marketed almost exclusively around software development when its underlying capabilities seem well suited to creative production workflows like this. Tibo, if you happen to see this: I'd love to see the Codex team explore and showcase more unconventional use cases, especially music composition, MIDI generation, audio tooling, and creative workflows built around GitHub repositories. I'm curious whether anyone else here has been experimenting with Codex for music production or other creative applications beyond writing software. Would be interested in comparing approaches.