File operations at scale: sort, rename, deduplicate

If you want to show someone what Cowork can do in under five minutes, run a file-organisation task on a messy Downloads folder. It lands every time — not because it's the most sophisticated use case, but because almost everyone has one, and the result is immediately visible.

The reason it lands is that Cowork is content-aware, not filename-based. When it processes your Downloads folder, it doesn't look at file names and sort by extension. It opens each file, reads what's inside, and makes decisions from the content. IMG_4521.png isn't categorised as "image" — it's recognised as a Stripe invoice and moved accordingly. A .pdf with "AGMT" in the name isn't filed under miscellaneous — Cowork reads the first page, identifies it as a contract, and places it with the contracts.

This is what separates Cowork from a traditional file-manager script. Scripts pattern-match on names and extensions. Cowork understands documents.

What content-aware file handling actually means

When Cowork reads a folder, it works through each file systematically: opens it, reads the content, infers what the document is, extracts any dates or metadata it can find, and decides how to categorise and name it. For images, it uses EXIF metadata to pull the actual capture date, then renames with a YYYY-MM-DD prefix so your photos sort chronologically regardless of what your camera named them.

The practical use cases:

• Downloads decluttering — sort 200+ accumulated files into logical subfolders based on what they actually are, not what they're called
• Photo organisation — rename with YYYY-MM-DD prefix from EXIF data, optionally grouped by event or trip
• Duplicate detection — finds near-duplicates (slightly different versions of the same document), not just exact-hash matches
• Batch format conversion — docx to PDF, image compression, format normalisation across a whole folder at once
• Archival by year, project, or client — structure your reference folders without touching each file manually

The sample prompt

Run this verbatim on a test folder and you'll understand exactly what Cowork does:

"Organize this folder into logical categories based on file contents. Rename files with YYYY-MM-DD prefix where dates are identifiable. Create subfolders that make sense. Log every decision in ORGANIZATION-LOG.md showing before/after for each file. Do not delete anything — move uncertain files to REVIEW/."

Four things make this prompt effective. First, it describes an outcome ("logical categories"), not a procedure. Second, it specifies the renaming format so Cowork doesn't invent its own. Third, it asks for a log — which means every decision is auditable. Fourth, and most importantly, it includes the safety clause: do not delete anything, with a designated REVIEW/ folder for anything uncertain.

That safety clause is not optional. Always include it on file-operation tasks until you've fully reviewed the output and are confident in Cowork's judgement on your specific folder. You can always delete from REVIEW/ later. You can't un-delete a file that Cowork misclassified.

Reading the log

The ORGANIZATION-LOG.md file Cowork creates is where the actual value lives. Every moved file gets a before-and-after entry with a reasoning note explaining why it was categorised that way. Read it. Not skimming — read it. You'll spot misclassifications, learn what Cowork found confusing, and understand exactly what happened to your files. The log is your audit trail, and it's the input for your next steering instruction if anything needs correcting.

The steering loop

If you see an error in the log — say, a batch of meeting notes filed under "Research PDFs" — correct it in the same session:

"Move all files with meeting notes content to a subfolder called meetings/. Update the log."

Cowork will adjust. This is the pattern: let it do the bulk organisation, read the log, correct the exceptions. You're reviewing at the category level, not the file level — which is exactly the leverage you want.