Level 3 · 2 to 3 hours

Build your AI personal assistant

You will design a small personal assistant you can reuse on a normal week, not a chatbot that pretends to know your private life. Start with a context card of public facts only: your role, the tasks you repeat, who reads your notes, and the limits you will not cross. Turn that card into a standing brief Gemini can follow for planning, short drafts, and decision notes. Practice one real task from this week, keep the prompt that worked, and add a check you perform yourself. Client names, secrets, and unpublished numbers stay out of the chat. You leave with a brief you control and one piece of work you edited and would actually use.

Outcomes

A context card of public facts that a fresh chat can follow without your old transcript.
A standing assistant brief for planning, short drafts, and decision notes, with clear refusals.
One real task from this week, edited by you after a check you can name.

Steps

1. Write the context card
List only facts you would leave on a shared desk: your role, three repeated tasks, who reads your notes, and what a good result looks like. Leave out other people's names, account numbers, health details, and unpublished figures. Keep the card under 120 words so a fresh chat can use it without dragging in an old transcript.
2. Set the assistant rules
Tell Gemini which tasks it may help with and which it must refuse, including sending mail, inventing meetings, and guessing private facts. Require one clarifying question when the reader, the task, or the length is missing, then one draft. Save these rules beside the context card so the next chat starts clean instead of inheriting a messy thread.
3. Run one real task
Pick a task you already owe this week, such as a short agenda, a reply outline, or a decision note with two options. Paste the context card and the rules, and name the reader, the length, and what done looks like. Ask for a draft only. You still decide what is sent, filed, or dropped after you read it.
4. Check the draft and edit one line
Read the draft against three checks you write yourself: a missing fact only you would know, a tone that does not sound like you, and a detail that was invented. Change at least one line in your own words before you keep the result. If the draft is unusable, add the missing constraint and run one more pass.
5. Save the reusable brief
Copy the context card, the rules, and the prompt that worked into a note you control. Add one line naming what you will never paste, such as secrets or unpublished numbers. Title the note so you can find it on Monday. A chat history is not a filing cabinet, and this academy does not sync that note for you.

Starter prompts

Context interview

Interview me so I can build a personal assistant brief I will reuse. Ask one question at a time and stop after five questions. Cover my role, the tasks I repeat, who reads my notes, what a good result looks like, and what you must not invent. Do not ask for secrets, passwords, names of other people, health details, or unpublished figures. If I say skip, move on without pushing. After my last answer, write a context card of at most 120 words using only what I chose to share. Then list two tasks this card can support this week. Do not draft those tasks yet.

Standing assistant brief

Use the context card I will paste under this instruction. Act as my drafting partner for planning, short replies, and decision notes. If the reader, the task, or the length is missing, ask one clarifying question and then stop. When those are present, produce one draft only. Do not invent meetings, names, prices, or dates I did not give you. Do not claim you sent, filed, or approved anything. End with three checks I should run myself: a missing fact, a tone mismatch, and a possible invention. Keep the draft short enough that I can edit it in one sitting.

Checklist

The context card uses only public facts and stays under 120 words.
The rules name what the assistant may draft and what it must refuse.
One real task from this week was drafted for a named reader and length.
I changed at least one line myself after a three-point check.
Secrets, other people's names, and unpublished numbers stayed out of the chat.
The reusable brief is saved in a note I control, not only in chat history.

Rubric

A fresh chat could follow the brief without your old transcript.
The finished task is something you would actually use after your edit.
Limits are explicit: no invented facts, no private data, and no claim that the assistant acted for you.
You can say, in one sentence, what you checked and what you changed.