Prompt To Session

Prompt to Session

A proof of concept for prompt-driven DAW automation using ChatGPT, Claude Cowork, Codex, and Ableton Live 12 Lite. AI agents generated MIDI, operated the DAW, diagnosed errors, repaired project state, and validated the export—demonstrating a human-in-the-loop workflow rather than autonomous music production.

Overview

This project explores agentic automation inside a professional digital audio workstation. The workflow began with ChatGPT, which was used to develop the initial prompt and phased production plan. Claude Cowork and Codex then took turns executing that plan inside Ableton Live 12 Lite.

The agents configured the project, imported structured data, created and repaired automation, diagnosed routing and MIDI-mapping problems, inspected the underlying project structure, measured output levels, managed project assets, and produced a technically validated export.

The purpose of the experiment was not to produce a release-ready track. It was to investigate whether prompt-driven AI agents could meaningfully participate in a complex, stateful desktop production workflow.

Reproducing the Workflow

If you want to run the same prompt-driven process yourself rather than just inspecting the finished project, see [`docs/HOW_TO_USE.md`](docs/HOW_TO_USE.md). It covers copying the production plan in [`docs/`](docs/) and pasting it into an AI agent with desktop control of Ableton Live, along with tips on supervising the session, handling interruptions/handoffs, and validating the export.

Tools Used

• ChatGPT — initial prompt development and production planning
• Claude Cowork — agentic execution and DAW interaction
• Codex — agentic execution, diagnosis, repair, and validation
• Ableton Live 12 Lite — target production environment

Hybrid Automation Approach

During execution, the agents determined that entering large quantities of MIDI notes manually through the DAW interface would be too slow and unreliable.

The agen