AI work notes
Leave a return path for every file you move
How I separate file cleanup into read-only inventory, a proposed path map, and small moves so classification involves less guessing.
Inspect before moving
When I helped a PM organize a folder, the hardest part was not the file count. Names like FINAL, FINAL2, and truly-final looked like riddles, and miscellaneous folders contained more miscellaneous folders. His concern was reasonable: if AI moved everything at once, would important files disappear from view?
For the first round I allowed only inventory: the tree, names, modification times, and possible duplicates in a named subfolder. I explicitly prohibited moves, renaming, deletion, and overwriting. Matching names and newer dates are clues, not enough to decide what to keep or discard.
Make a path mapping table
Only in the second round did I ask for original path, proposed destination, reason, and risk. I check whether the categories match how I will find things later. Project and purpose may be enough; files with unclear purposes stay for human review.
Take two presentation files as an example. I retain the full paths and inspect contents before deciding how dates and purposes belong in their names. I want AI to reduce the effort of proposing categories; version decisions still need evidence from the files.
Check conflicts before a move
After reviewing the map, I check that destinations stay in the named directory, look for collisions and references, and verify backup or recovery options. For program and project folders, I check dependencies first. A tidier structure can otherwise break something that was working.
For synced folders, I confirm files are present locally. Permission, sync, or path problems stop the move until resolved. I schedule moving and deleting separately, even when files appear duplicated, so one batch does not combine two different risks.
Move small batches with a log
I used a maximum of twenty files per batch; an unfamiliar folder can start with fewer. Each entry records its original path, actual destination, result, and failure reason. I stop at collisions or permission problems instead of overwriting. A small batch makes the affected files easier to identify.
After each batch I reconcile counts, open sample files, and check links I use. I continue after verification. Recovery reverses checked movement-log entries. I leave deletion for another review round rather than combining it with moving files.
Keep rules you can maintain
At the end I keep a few maintainable rules: where new files go, when to archive, and what to do with unclear purposes. Names need only information that helps retrieval. If the old name carries useful context, I preserve it in the mapping table.
The value is less guessing when I look for files and a clear home for new ones. I get one small folder working before expanding. Traceable moves, a stopping point at conflicts, and a return path make the routine easier to keep using.