Build your personal AI assistant

Before building anything, let's start with the basics.

What is an AI assistant?

An AI personal assistant is a customised digital partner that helps you organise, filter, and plan your daily professional workload.

Ideally, an AI personal assistant operates inside a human-first context engine. A human first context engine is designed to curate work behaviour, role defination and goal oriented. It holds permanent knowledge of your specific job role, your work objectives, and your shifting priorities.

Why you should have one

Life is busy and we get information from many different sources, making prioritasation a new task. Therefore, a dedicated AI assistant solves the three biggest friction points of modern life:

  • It Cures "Where do I start problem": Starring at a blank page or being overwhelmed with information is something we have all experienced. With and AI assistant who already knows your professional baseline and goals, you to simply drop in raw notes, meeting transcripts, or inbox data and immediately ask for a strategic next step.

  • It Stops Generic AI Text": A well configured and designed AI assistant uses your specific personal context to enforce strict communication guardrails, that aligns with your real voice and specific needs.

  • It Bridges the Human Adoption Gap: The ultimate bottleneck in technology adoption isn't access to software; it's consistency and trust. Having a customized AI assistant integrates into your actual daily routines, moving you from passive experimentation to active, repeatable AI Fluency.

How to build the AI assistant

Before you build, this is what you need

  1. Your computer:
  2. A paid subscription for your prefered LLM Client (for instance Claude or ChatGPT)
  3. Time. Set aside atleast one hour to do this.

Note: This assistant build flow is an adaptation of Reframe Tech’s previously held live build session. Treat is as the shared companion to the build and updates may have been made within the interface of the tools mentioned.

Step 1: Teach it who you are

This is what makes your AI assistant YOURS. Create a folder on your computer or cloud (eg google drive) and name it my AI assistant.

  • Paste the prompt below in your prefered LLM Client. Let it interview you, one question at a time.

    I want to create an about-me.md file for my AI assistant. Please ask me the following questions one at a time, and wait for my answer before moving to the next one:

  1. What is my current role or situation?
  2. What does success look like for me in the next 12 months?
  3. What kinds of opportunities am I looking for? (eg jobs, business, fractional work, mentorship, events, volunteering, etc)
  4. What is my communication tone? (e.g. "bullet points only, active voice, neutral tone, avoid corporate jargon"). I don't want to sound like AI.

When we have been through all five, compile my answers into a clean markdown file with the heading "About Me" formatted with clear headings and one section per question. Start with the first question now.

  • Save the answers as about-me.md in your folder.

Pro tip: The honest version works better than the impressive version.

Step 2: Tell it what to do

This will be your standing instruction that will do five jobs: inbox scan, calendar prep, opportunity surface, signal scan, and the hidden opportunity radar.

Copy the instruction below, don't type it.

Instruction:

You are my personal assistant. Every morning, do the following and give me one clear brief. Start by stating today's date.

Rules for the whole brief: Each item appears in only one section; put it where it's most useful and don't repeat it. If a section has nothing worth flagging, say "Nothing today" and move on. Never invent or pad; only surface things actually in my inbox, calendar, or about-me.md.

  1. TOP OF THE LIST: The single most important thing I should do today, in one sentence, before anything else.

  2. INBOX SCAN: Search my email for messages from the past 24 hours. Flag anything that needs a reply, comes from a person (not a newsletter), or mentions an opportunity, deadline, event, course, or application. For each one give me: sender, subject, a one-line summary, and the action I should take. For the two most important emails that need a response, draft a reply I can review; keep each to 3–4 sentences as a starting point, in the tone described in about-me.md.

  3. CALENDAR PREP: Check my calendar for today. For each appointment or meeting: what it is, who is involved, and one thing I should prepare, say, or ask.

  4. OPPORTUNITY SURFACE: Look at my inbox from the past 7 days. Flag any replies I need to make, follow-ups I have missed, invitations I have not answered, or conversations I should restart.

  5. SIGNAL SCAN: Read my about-me.md file from my connected folder. Then scan my inbox from the past 48 hours for anything that matches where I want to go: events, courses, articles, introductions, or news about organisations related to my goals. Flag ONE thing I should act on this week and tell me why.

  6. HIDDEN OPPORTUNITY RADAR: Cross-reference my about-me.md against everything in my inbox and calendar from the past 48 hours. Read between the lines and show me what I would not see on my own. Only include items genuinely present in my inbox or calendar; if there's nothing real, say so rather than inventing:

  • A person I should reach out to this week, and a draft opening line in my voice.
  • Someone who mentioned a topic, event, or challenge that connects to my goals. Tell me who, what they said, and exactly how I could reply.
  • An introduction or piece of help I should ask for, based on who I am already talking to and where I want to go.
  • Something happening in my target field that I should know about or have an opinion on. Give me the short version.

Format: Clean sections with clear headings. Plain language, no jargon. I read this in the morning before my first coffee. Tailor every section to what you find in about-me.md.

Pro tip: Subscribe to 2-3 leading newsleters in your industry to enhance steps 4 (Signal scan) and 5 (Hidden opportunity radar)

Step 3: Set it up

On Claude

Open Claude on desktop or browser → Projects (left sidebar) → New Project (popup appears):

What are you working on? (Give it a name for insstance: "My personal assistant" What are you trying to achieve? (Paste in your master prompt) Context (point it to the folder you created earlier) Click "create project" (congratulations you assistant is setup, now make it work for you).

On left panel

You can see what you have done so far:

  • Instructions

  • Context

  • Memory (If enabled, it allows Claude to share memory between this project and your chats outside of projects). You decide if you want to enable it or not.

  • Scheduled (Let's set this up)

    Click "Add"
    - Name it: "My morning brief"
    - Description: Summary "My morning brief" (Paste your master prompt into the description box)
    - Auto: (This is about permissions: choose based on your level of comfort: Ask before acting, or Act without asking)
    Model: Decide based on your preference: Eg. Opus 4.8
    - Trigger: Decide based on your preference: Eg. Daily: 08.30
    - Advanced settings: (We'll set it up in the next step) For now click "create" to leave this window.
    

On the right panel

Go to customize → Connectors (find and connect):

  1. Email (your prefereed email, eg. GMAIL)
  2. Calender
  3. Drive
  4. Any other tools you want your assistant to have access to.

Congratulations now you are all set!

On ChatGPT

Open ChatGPT on desktop or browser → Projects (left sidebar) → New Project (popup appears):

Project name: (Give it a name for insstance: "My personal assistant"

Sources/ Add sources: (point it to the folder you created earlier)

Click: "create project" (congratulations you assistant is setup, now make it work for you).

Connect plugins

Plugins on ChatGPT is the way to connect your work across your tools.

Go plugins → Connect plugins/browse all plugins (find and connect):

  1. Email (your prefereed email, eg. GMAIL)
  2. Calender
  3. Any other tools you want your assistant to have access to.

Congratulations your assistant now has context.

On left panel

Click scheduled (clock icon) → New task (Schedule a task pops up):

  • Instrustions: (Paste your master prompt into the description box)
  • Repeat: Daily
  • Time: 08.30
  • Advanced: (Configure according to your preferences)
    • Run on this computer: (Toggle on or off)
    • Start each run in new chat: (Toggle on or off)
    • Model: "GPT-6 Astra"
    • Effort: Choose level: "light, medium, high, extra high, ultra"

Click Create

Congratulations your assistant will now run every morning.

Addiotionally, Go to your project "My personal assistant"**

In the chat, paste your master prompt and ask it to be linked to your scheduled task.

Congratulations you are now all set!

To test the entire setup, go the project "My personal assistant" in the chat, ask: "what are my latest 5 emails," "what's on my calendar for tomorrow?" "run my scheduled task."

Take 2 minutes to review the output. If the response feels generic or misaligned, immediately update your Personal Context Document (this is about me.md file in your connected folder) to refine the engine's long-term memory.

Pro Tip: Your assistant should grow to become part of your system. Once you feel ready, you can upgrade it by creating Skills on Claude. Or Skills on ChatGPT.

Keep the Momentum

This assistant build flow is an adaptation of Reframe Tech’s previously held live build session. Join the next session by signing up to our next event. See the calendar below and how to signup. We limit access to 15 people to ensure deep, focused troubleshooting and learning.

Building an AI fluency movement through practical workflows, collaborative AI builds, lunch-and-learns, podcast events and exploring the future of work together.

Come as you are. Leave a little more AI fluent.