When should I use an Agent?
20 October 2026
Learn how to identify tasks that are suitable for an AI agent, and where people should remain in control.

Today's Lesson
An AI agent is an AI system that works towards a goal across multiple steps. But not every task needs an agent.
Should you use an agent for your task? Let's use 4 simple checks to assess:

📝 We will use the example of preparing minutes after a weekly team meeting to illustrate this.
1. Does it involve multiple steps?

✅ Agents work well for tasks that involve multiple steps, for example:
Retrieve the transcript
Identify the key discussion points
Prepare the minutes in the required format
Review the minutes for accuracy
Circulate the minutes
This makes it possible to define a sequence of work for an agent to follow.
2. Is the task repeatable?

✅ Agents work well for processes that repeat regularly, such as generating minutes for a weekly meeting.
Instead of explaining the task from scratch each time, you can design a reusable workflow for the agent to follow.
3. Does it need information or tools?

✅ Tasks that require information from connected sources, or tools to take action are good candidates for agents. In this case the agent will use:
Information: Identify the meeting using Outlook Calendar, retrieve the transcript using Microsoft Teams.
Tools: Generate a Word Document in the required format, send the approved minutes using Outlook Email.
4. Is the output reviewable?

✅ A suitable agent workflow should have clear points where a person can check the output or approve actions before they proceed.
Here, a person can compare the AI-generated minutes against the meeting transcript to check that they are accurate before approving it's circulation.
Putting it together
Choosing the right use case is not just about whether to use an agent. It is also about deciding where the agent's responsibility ends and the person's responsibility begins.

Now it's your turn to try!
Think of a task you do regularly and put it through the same four checks.
Multiple steps: Does completing the task require several actions?
Repeatable: Does the same workflow happen regularly?
Information or tools: Does it need information from other sources or access to tools?
Reviewable: Is there a clear output that someone can check?
If you answer yes to these questions, the task may be worth exploring as an agent workflow.
