You are doing work it could structure

Why this works

If you can state the rule ("each of these becomes a row with these columns"), the model can apply it across all of it at once — reformatting, extracting, renaming, converting between formats, turning notes into a table. Your time then goes where it is actually needed: reviewing the output, which is faster than producing it and catches the two rows where the rule was ambiguous. The "mark ambiguity instead of guessing" instruction is what keeps the review honest.

When this applies

You notice you are doing the same small transformation the third time, and you could explain the transformation to a colleague in one sentence.

Example

Instead of Retyping figures from a PDF report into your own summary.

Try From the attached PDF, pull every figure in section 4 into a list: metric, value, page number. I'll spot-check against the pages.

Template

Here is the raw material: [paste or attach]. Turn each [unit] into [target shape with named fields]. Where the material is ambiguous, mark it "?" rather than deciding.

Go deeper

Files and PDFs have their own mechanics — what actually gets read. When the same transformation recurs weekly, that is a job for an agent: what Cowork is, and for the honest question first, do you need an agent at all. The spot-check habit is non-negotiable for numbers.