Clef Sheet Music Editor

About this project

In this experiment, I tested Claude Code's ability to work in a more agile-based way, with shifting or growing requirements, instead of a one-shot way (which is how I saw a disproportionate number of experiments online being attempted). The theme I chose was a browser-based sheet music transcription application, as I am familiar with similar tools like MuseScore, Finale, and Sibelius. Consistent with the spirit of this experiment, I would create an MVP sheet music transcription application and then iteratively ask for more and more features to be added. In the end, only a few iterations were performed, and it taught me how Claude Code performs in this type of experiment compared to a one-shot.

I found that Claude Code successfully added the new features that I requested. One interesting thing is that Claude Code seemed hesitant to change what previously existed in the code, even if refactoring a portion would result in fewer files or lines of code with the same functionality. This can potentially be seen as a good thing in general: if Claude Code is working on an existing baseline, it should try to keep that baseline as similar as possible after an incremental change, so as not to introduce unanticipated behavior or refactor something that is already familiar to developers. However, in a project like this, where Claude Code has full control over the code, a better strategy in the future could be to encourage Claude Code to perform refactors, as long as it is confident in them and has a thorough test suite to ensure there is no drift in capabilities or introduction of bugs within these refactors.