patrickz.aiLet’s talk ↗

Automate safely / Guide

Plan a safe AI home assistant, starting with one lamp

Design one safe AI command for an imaginary lamp: the AI proposes, a simple rule decides, and failure is honest.

About 30 minutesSome experienceRead free · No sign-up

Before you start

Only a chat assistant. You will simulate two imaginary lamps, a desk lamp and a bedroom lamp; no smart-home devices or code needed.

Why this lesson exists

The local home-assistant experiment kept two jobs apart: the AI works out what you meant, and a separate, simple check decides whether anything is allowed to happen. Practice that split with an imaginary lamp before any model gets real control.

The idea this guide practices: Copilot Cowork tutorial: build and schedule your first skill.

Do the exercise

  1. List what is allowed

    Write the only actions allowed: turn the desk lamp or the bedroom lamp on or off. Anything else is not allowed.

  2. Test five requests

    Paste the prompt below, then send: “Turn off the desk lamp”, “Lights on”, “Turn on the bedroom lamp”, “Set the kitchen to 50%”, “Make it cozy”. Note which it proposes, asks about or refuses.

  3. Check each proposal yourself

    Compare every proposed action with your allowed list by hand, the way a separate check would. Only an action on the list counts; the AI saying “Done” does not.

  4. Plan for failure

    Write what a person sees when the lamp or the assistant fails, and make sure your plan never removes the normal wall switch.

A prompt to adapt

Replace the bracketed parts with your own details.

You are planning commands for an imaginary home with two lamps: [desk lamp, bedroom lamp]. You may only propose one action: turn a listed lamp on or off. For each request I send, reply with the lamp and on/off you propose, ask one question if it is unclear, or say it is not allowed. Never say an action happened; you can only propose it.

More prompts to adapt ↗

What this looks like

Asked “Make it cozy”, a safe assistant asks which lamp instead of inventing a dimming level. Asked about the kitchen, it says there is no kitchen lamp on the list.

Check your result

Use evidence from your output. A confident explanation from the AI is not enough.

If it isn’t working

If the assistant keeps saying Done, repeat the rule that it may only propose. Running the AI on your own computer does not keep things private if the lamp itself talks to a company’s cloud. Before connecting any real device, list every app and service the command passes through.

Where this came from

Adapted from the archived LinkedIn theme “I Gave My House a Brain and Refused to Give It the Internet.” See the post coverage and editorial method.

Prepared September 2026. Tools and interfaces change; use current official setup instructions. Session lengths are estimates.

THE IDEA BEHIND THIS GUIDE

Read the story, then keep practicing

KEEP GOING

Your next useful step

Browse all 57 guides ↗