All Citizens classes
3percentclub

Citizens: Week 3

Let AI do tasks for you, safely

What changes when AI stops just talking and starts doing things, and how to stay the one who presses send.

Flip through the slides: "Don't Let the MCPs Turn You Into an NPC"

The big ideas

  • Talking vs. doing

    A chatbot that gives a wrong answer is annoying. An AI that sends the wrong email, books the wrong date or deletes a file is a real problem. Doing raises the stakes.

  • You hold the send button

    Let AI draft and prepare. Anything that spends money, contacts people or can't be undone waits for your yes.

  • Four levels of AI agency

    From a textbox you type into, to a helper that suggests, to an assistant that acts when you approve, to a background runner that acts on its own. Know which level you're giving it.

  • Skills are reusable recipes

    Instead of retyping the same prompt, write it down once as a recipe: the goal, the steps, the rules and what good looks like. No coding needed.

  • Connecting your apps, and 4 questions first

    Connectors (like MCP) let AI reach your email, calendar or files. Before you connect anything, ask: what can it see, what can it change, who approves, and how do I undo it?

Try it yourself

  • Write down one prompt you've typed more than three times. Turn it into a recipe.
  • Open the connected apps list in your AI tool and ask the 4 questions for each one.
  • For one task you'd hand to AI, mark the exact step where it has to stop and ask you.

Go deeper: the class readings

Homework

Week 3 Homework: The Food Pantry Helper

Level 1 and the reflection questions are the full assignment (about 20 minutes). Levels 2, 3 and the bonus are optional challenges. Copy the worksheet into your own doc and write your answers under each level.

The situation

Dee runs volunteers at the Elm Street Food Pantry. She wants an AI assistant connected to the pantry's shared email, calendar and volunteer spreadsheet to help with the Saturday shift.

What Dee wants the assistant to do

1. Read new volunteer sign-up emails

2. Add new volunteers to the spreadsheet

3. Draft a welcome email to each new volunteer

4. Send the Saturday shift reminder to everyone on the list

5. Move a volunteer to a different shift when they ask

6. Remove volunteers who haven't come in 3 months

  1. Level 1Required15–20 min

    Let it or stop it?

    1. For each of the 6 tasks, write LET IT (AI can just do it) or STOP IT (AI prepares it, Dee approves).
    2. For each task, write which agency level it is: textbox, suggests, acts with approval, or runs on its own.
    3. Pick the one task that worries you most and explain in 2–3 sentences what could go wrong.

    A complete answer

    • All 6 tasks marked LET IT or STOP IT
    • Sending to everyone (4) and removing people (6) are STOP IT
    • Each task has an agency level
    • The worry names a specific thing that could go wrong for a real person
  2. Level 2Optional15 min

    Ask the 4 questions

    1. Dee wants to connect three things: email, calendar and the volunteer spreadsheet.
    2. For each one, answer: what can it see, what can it change, who approves, and how do we undo it?

    A complete answer

    • All three connections covered
    • Noticed the spreadsheet holds volunteers' phone numbers and addresses
    • Each one has an undo plan
  3. Level 3Optional15–20 min

    Write the Saturday reminder recipe

    1. Write a skill recipe Dee could reuse every week for the shift reminder.
    2. Include: the goal, the steps, the rules (what it must never do), and what a good reminder looks like.

    A complete answer

    • Has goal, steps, rules and a good example
    • A rule says Dee approves before anything sends
    • Written in plain language anyone at the pantry could follow
  4. BonusOptionalOpen-ended

    Try it with a real assistant

    1. Paste your recipe into an AI tool (no real names or contact info) and ask it to draft this week's reminder.
    2. Write two sentences: did it follow your rules? What would you change in the recipe?

    A complete answer

    • Used made-up names only
    • Named one change to the recipe

Reflection (required)

  1. Where in your own life would you let AI act on its own, and where would you never?
  2. If the assistant made a mistake on task 6, who would be hurt and how would they find out?
3percentclub · AI Architect Fellowship