The AI Work Starter: 10 Tasks to Try First
A short list of low-risk tasks that show you what AI is actually good at in ordinary work — and where it is not worth the trouble.
Most people stall with AI because they start with the wrong task. They try something high-stakes, get a mediocre result, and conclude the tool is overrated. A better approach is to begin with tasks where you already know what a good answer looks like, so you can judge the output immediately.
The ten tasks below are ordinary work, not demonstrations. Each takes a few minutes and gives you a clear read on whether the result was useful.
The ten tasks
- 01Summarize a long email thread into the decision, the open questions, and who owes what.
- 02Turn rough meeting notes into a structured recap with action items and owners.
- 03Draft a first version of a routine message you write often — a status update, a follow-up, a scheduling note.
- 04Rewrite something you already wrote to be shorter, or clearer, or less blunt.
- 05Explain an unfamiliar document — a contract clause, a technical spec, a policy — in plain language, then ask what it does not say.
- 06Build a checklist for a process you run repeatedly, then correct it from your own experience.
- 07Compare two options you are weighing, and ask for the strongest argument against the one you prefer.
- 08Convert unstructured information into a table: vendors and criteria, candidates and skills, tasks and effort.
- 09Prepare for a conversation by listing the questions the other person is likely to ask.
- 10Draft an outline for a document you have been avoiding, so you are editing instead of starting.
How to judge the result
For each task, ask one question: did this save me time without lowering the quality of my work? If it saved time but you had to rewrite everything, the task is a poor fit or your instructions were too thin. If the quality held and you moved faster, keep it.
Write a status update for my project.
Write a five-sentence status update for a weekly leadership email. Project: vendor migration. Done this week: data mapping finished, two integrations tested. Blocked: security review not scheduled. Next week: begin pilot with the finance team. Tone: factual, no adjectives.
What to skip at the start
- Anything where you cannot verify the facts yourself.
- Work with confidential information your employer has not cleared for AI tools.
- Final-form work that goes out without review.
- Tasks where the whole value is your personal judgment or relationship with the reader.
What to do next
- Pick three tasks from the list and run them this week on real work.
- Keep a short note of which ones held up and which ones did not.
- Read the AI Output Review Checklist before you send anything AI-assisted to someone else.