How to Start Using AI at Work: Choose One Task You Can Check
A practical starting worksheet for people and teams new to AI: choose one task, define the result, keep a person responsible, and check whether the experiment helps.
If you want to start using AI at work, begin with a task you understand. You need enough context to recognise a useful result and notice when something is wrong.
You don't need to redesign your whole working day. Pick one small piece of work, decide what AI is allowed to do, and compare the result with how you handle it now. Treat it as an experiment, with permission to stop.
Choose a task before choosing a tool
Write down three tasks you repeat. For each one, ask:
- What makes this frustrating: collecting information, preparing a draft, checking details, or waiting for someone?
- Do I have reliable information to start from?
- Can someone check the result before it affects anyone else?
- Could a template or a clearer process solve the problem already?
For a first trial, I'd choose work where AI prepares something for review. Keep sending, publishing, spending, and changing records with a person. If mistakes are difficult to spot or undo, choose a simpler starting point.
I explain this decision in more detail in what to automate, where, when, and how much.
Fill in this worksheet
Before opening a tool, write short answers to these questions:
| Decision | What to write down |
|---|---|
| Task and reason | What happens repeatedly, and what would you like to improve? |
| Input | What information will AI receive? Is it complete, and are you allowed to use it in the chosen tool? |
| Output | What exactly should come back: a draft, a list, a comparison? |
| Boundaries | What must AI leave unchanged or ask about? What actions stay with a person? |
| Check | How will you compare the result with the source and spot missing or invented details? |
| Reviewer | Who understands the task and decides whether the result is usable? |
| Cost and value | What will you record: preparation, review and correction time, tool cost, and mistakes? |
For workplace information, follow your organisation's rules and confirm which tools and data are permitted. If that is unclear, use invented example material while you resolve it.
A worked example: preparing an event email
This is a hypothetical example, not a client project or a report of results.
Imagine a workshop organiser who repeatedly writes joining instructions. The aim is to prepare a draft from approved event details. AI does not send the email.
Input: The approved public date, time, venue, booking link, and what attendees should bring. No attendee list is needed for this trial.
Output: A short email with a subject line and a checklist. Ask it to flag missing information instead of filling gaps.
Check: Compare every date, address, link, and instruction with the source. Check that the draft has not added a promise, changed the schedule, or left out something attendees need.
Reviewer: The organiser, who owns the event details and can reject the draft.
Cost: Record the time spent preparing the input, checking the email, and correcting it, alongside any tool cost. Compare that with writing a similar email manually. A quick first draft is only one part of the work.
Decide what to keep
Try a small set of examples, including one with missing information. Note what needed correction and whether you could reliably catch it. If reviewing takes longer than the original task, simplify the process or choose another experiment.
Don't increase AI's control just because the first result looks good. Keep checking the result, even after successful runs.
Want help choosing your first task or reviewing the worksheet? Read about my AI adoption and workflow automation advice. Bring the task, what makes it difficult, and what you are unsure about.
