Practical AI
No. 4.05Track 4 · Build small tools

AI agents explained simply: plan one on paper

Lesson 5 of 55 min read1 task · 3 questions

An agent is a helper that works in steps. Plan the steps before you trust it.

The idea

An AI agent is a chatbot that can take several steps toward a goal and use tools, such as search, email or a spreadsheet, along the way. A normal chatbot answers once. An agent keeps going until it thinks the job is done.

That power needs limits. Every good agent plan has five parts: Goal, Tools it may use, Steps, Stop rule (when it must stop) and Human check (what you approve before anything is sent or deleted).

Start on paper. If you cannot write the steps for a person to follow, an agent will not manage them either. Start with read-only jobs, such as sorting and summarising. Add actions like sending or deleting only after you trust it.

Your task

Plan a weekly inbox clean-up agent

You will not build it yet. You will write the plan a safe agent needs.

Your slip: copy-paste prompt
Help me design a safe AI agent plan, on paper only, for this job: "Every Monday, go through my unread work emails from last week and give me a short list of what needs a reply."

Write these five headings with short answers: Goal, Tools it may use, Steps (max 5), Stop rule, Human check.
Safety rules to build in: it may only read and summarise. It must never send, delete or move any email. It must not open attachments or links.
Keep it under 120 words.

What you should get

A real example of a good result. Your wording will differ a little. The shape and the facts should match.

Printed resultExample

Goal: Give me a short Monday list of emails from last week that need a reply.

Tools it may use: Read access to my unread emails. No other tools.

Steps:
1. List unread emails from the last 7 days.
2. Skip newsletters and automatic notifications.
3. Mark each remaining email as Needs reply, Needs action or Just FYI.
4. Write one line per Needs reply email: sender, subject, what they ask.
5. Show the list, newest first.

Stop rule: Stop after one pass. Stop and tell me if more than 100 emails.

Human check: I read the list and reply myself. The agent never sends, deletes or moves anything.

Check yourself

Q1 What is the main difference between a chatbot and an agent?

Q2 What is a stop rule?

Q3 Which first agent job is safest?

Finished?

Stamp this ticket. Your pass on the home page gets punched.

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5