Level 2 · Power user
Agents: when AI acts, not just answers
An agent gets a goal, plans steps and uses tools.
A chat answers you. An agent works: it gets a goal, splits it into steps, uses tools like your email or calendar, checks the result and keeps going until it's done.
It's like going from an assistant that answers questions to one you give a mission. So the key question is no longer just "what does it know" but "what can it do without asking you".
Try it today
Pick a boring, repeated task from your week. Write it step by step as if explaining it to someone new. That's the draft of your first agent.
Test yourself
Which step should a new agent never skip?
- Asking for confirmation before sending or deleting
- Using emojis
- Working at night
See the answer
Asking for confirmation before sending or deleting. Reading is low risk. Sending, paying or deleting can't be undone: a person goes there.
If you lead a team
Before approving an agent, ask three things: what it can do alone, what needs approval and who reviews its logs.