Automating Repetitive Busywork
Find the tasks worth automating and hand them to agents: inbox triage, file cleanup, form filling, and recurring reports.
Everyone has a list of tasks they dread: sorting the inbox after a week away, renaming 300 files from a client upload, copying details from PDFs into a spreadsheet, assembling the same status report every Monday. None of it is hard. All of it eats hours, and most of it is exactly the kind of work agents now handle well.
But "automate everything" is how people end up with an agent that emails a client the wrong invoice. This lesson shows you how to pick the busywork that's actually worth handing off, and gives you four tested playbooks (inbox triage, file cleanup, form filling, and recurring reports) built around one safety pattern: the agent proposes, you approve, then it acts. You'll finish with at least one chore permanently off your plate.
Find the tasks worth automating
Not all busywork is equal. Before you automate anything, score it on five questions:
| Question | Good candidate | Poor candidate |
|---|---|---|
| How often do you do it? | Weekly or more | Twice a year |
| Can you explain the rules in a paragraph? | "Invoices go in /finance/YYYY-MM" | "It depends on the client's mood" |
| How bad is a mistake? | Annoying, easy to undo | Expensive, public, or permanent |
| Can you check the result quickly? | A glance at a list or table | Only by redoing the work |
| Does the agent have access to the inputs? | Files, exports, pasted text | Locked inside a system with no export |
The sweet spot is frequent, rule-based, low-stakes, and quick to check. If a task scores badly on "how bad is a mistake," it isn't necessarily off limits, but the agent should only ever produce a draft for you to act on.
Run a busywork audit
Most people underestimate how much of their week is automatable because the chores are invisible to them. Let an agent help you see them:
Help me find tasks in my job that an AI agent could take over. Interview me about my typical week, one question at a time: what I do on Monday mornings, what I copy between tools, what reports I make, what I search for repeatedly, what I put off. After about 10 questions, give me a table of candidate tasks with: estimated hours per month, how rule-based it is (1–5), cost of a mistake (low / medium / high), and a suggested first step. Sort by hours saved for low-risk tasks first.