Audit your week before you hand the first task over

1 · Learn the move · Template pattern

A practice picks its first AI job from whatever the conference was excited about: a research tool, a notice-answering machine. Pick it from your own week instead. Write down what you did, how long it took, whether it repeats, and whether a client's return information was in your hands while you did it. Then sort every line. DRAFT: the model writes a first version from de-identified facts you pasted (the research memo skeleton, the plain-language notice translation, the month-end narrative). CHECK: it tests one thing against a source you pasted (the memo's citations against the list you opened, the trial balance against the close checklist, the letter against the firm's rules). YOURS: any position you sign, any figure that goes to the Service, any client conversation, anything that needs the client's name to be useful. DISCLOSURE: any line where the work cannot be done without return information going into a tool, which means it waits for a consent naming the tool or it stays by hand. The model sorts; it does not decide. A week audited this way hands over one DRAFT job first, and never a DISCLOSURE one.

Here is my week, one line per task with the time it took and whether client return information was in front of me: [paste, de-identified]. Sort every line into DRAFT (first version from de-identified facts I paste), CHECK (one thing tested against a source I paste), YOURS (I sign it, it goes to the Service, or it needs the client's name), or DISCLOSURE (cannot be done without return information going into a tool). For each DRAFT line say what I would paste and what I would have to open afterwards. Never move a DISCLOSURE line to DRAFT by assuming I will de-identify it; flag it and stop. End with the one DRAFT task to hand over first and why.