quartz saddle
Right now, the workflow is pretty frustrating: I upload a file, ChatGPT can read it and generate a modified version, but it cannot update the original file. However, it can keep creating new copies on every iteration.
In real work, nobody gets an Excel, Word document, or PDF perfect on the first try. The normal workflow is to review and correct several times. If every correction creates another file, the library quickly becomes full of versions like:
file.xlsx
file_updated.xlsx
file_updated_2.xlsx
file_final.xlsx
file_final_now_really_final.xlsx
That is not a useful library. That is an AI-powered document landfill.
A simple proposal:
- File-level permission: read-only / editable.
- Newly generated files: temporary (downloadable) / ask before saving to the library.
- Before overwriting: show the changes and ask for confirmation.
The library should not just be a storage place for frozen, duplicated, uncontrolled files. It could easily become a trash container. It should work more like a document workspace: open, edit, review, save, version, or discard.
Also, having the potential to work directly with user files without depending on third-party tools, and not taking full advantage of it, feels like a product strategy mistake.