01 / Choose the task
Choose your first AI workflow.
Choose one small task you already understand. Start where you can explain the task, check the result, and undo a mistake.
A workflow is simply the steps from an input to a useful result. Meeting notes in, a draft action list out.
Which task should you start with?
Think of something you repeat and can check against material you already have. Keep the first attempt small enough to review in full.
| Task | Give it | Check it against |
|---|---|---|
| Draft an action list | One set of meeting notes | The actions, owners, and deadlines in the notes |
| Compare options | Two product descriptions | The stated features and your requirements |
| Organise a to-do list | Your tasks and stated deadlines | What is missing, duplicated, or scheduled unrealistically |
Use only material you’re allowed to share with the tool. A fictional example is enough to practise the method.
Can you explain it?
Name the input, the result, and the rules. “Help with this meeting” leaves too much open.
Turn these notes into a draft action list. Include an owner or deadline only when it appears in the notes. Flag missing details. Keep suggestions separate from agreed actions.
Can you check it?
Decide what would count as an error before asking AI to work. For meeting notes, a neat table still fails if it invents an owner or turns a suggestion into a commitment.
Illustrative example / Fictional meeting notes
“Maya will send the revised slides on Thursday. We need to ask the venue about step-free access; no owner was assigned. Leo suggested a rehearsal, but we didn’t agree to one.”
| Action | Owner | Deadline |
|---|---|---|
| Send revised slides | Maya | Thursday |
| Ask about step-free access | Not assigned | Not stated |
The human check: the rehearsal stays a suggestion. It isn’t an agreed action. The missing owner and deadline stay visible until someone supplies them.
Can you undo it?
Keep the result as a draft. Reviewing an action list is a useful first experiment. Automatically sending commitments to a client adds consequences before you know whether the output is reliable.
If the task is hard to reverse, narrow it to preparation. You review what happens next.
How do you write a reusable AI brief?
An AI task brief should name the input, expected output, rules, checks, and stopping point. If AI needs background, add your role, goal, audience, standards, and constraints. Keep that context relevant and update it when the work changes.
Task: [One thing I want done] Input: [The material to use] Output: [What the result should look like] Rules: [What must stay accurate or unchanged] Check: [How I will verify the result] Stop before: [Anything that needs my approval] Flag missing information. Don’t invent facts. Return a draft for me to review.Download the brief
How do you know whether AI saved time?
Compare the time the task normally takes you with the full AI-assisted attempt: preparation, generation, checking, and corrections. Use comparable tasks and the same quality standard. A fast draft alone doesn’t establish a time saving.
Keep the original beside the draft and record what you corrected. If the checked result is useful, try the brief on another example. If the full attempt takes more work than doing it yourself, narrow the task or keep it manual.
Hrishi Bhat builds AI workflows and teaches non-technical professionals how to use them, with results they can check.