Creating Your Own AI Assistant
Episode 10 shows how to create a simple, useful AI assistant with a clear role, practical instructions and a place to live, without needing technical skills or expensive tools.
What this episode is about
Episode 10 of Zero to AI is about creating your own AI assistant. Not a futuristic digital twin or a complicated automation system. A simple, useful assistant that helps with real tasks in your life and work.
The episode makes the idea practical. You do not need to be technical. You do not need to spend money. You need three things: a clear role, good instructions and a place for your assistant to live.
This matters because many people use AI as a blank chat box. A well-instructed assistant turns that blank box into something more dependable: a planning partner, writing helper, research reader or personal admin support.
An AI assistant is just a well-instructed chat with a purpose.
What an AI assistant actually is
People often imagine an AI assistant as a huge system that logs into everything, controls their calendar and manages their life. In this episode, the definition is much simpler.
A practical AI assistant has three ingredients: persona, tasks and home. The persona defines who it is and how it behaves. The tasks define what it helps with. The home defines where it lives.
Who the assistant is, how it behaves and what style of help it provides.
The specific jobs it helps with, such as planning, writing, research or admin.
Where it lives, such as ChatGPT, Gemini, Claude, Copilot or a reusable prompt document.
The goal is not a magical assistant. The goal is a focused helper that knows its role.
One assistant or many?
The big trap is trying to create one super assistant that does everything: writes, plans, researches, coaches, cooks dinner and probably fixes the printer. That usually creates confusion.
A better approach is to think like a manager. In a real team, you would not ask one person to be your accountant, marketing lead, lawyer and therapist. Different roles need different assistants.
Start with one assistant for one clear role. Once that is working, add specialist assistants for other jobs. Later, you can create a front-door assistant that helps you choose which specialist to use.
Examples of specialist assistants
Content Assistant
Helps draft emails, posts, scripts, summaries and content outlines in your preferred tone.
Research Assistant
Reads long documents, pulls out key points, compares options and highlights risks or opportunities.
Planning Assistant
Helps plan your week, break projects into steps and keep expectations realistic rather than fantasy-land.
Personal Admin Assistant
Helps write replies, tidy notes, turn chaos into lists and reduce the number of loose ends in your head.
Sales Support Assistant
Helps qualify leads, draft outreach, summarise calls and tidy CRM notes.
Where your assistant can live
You can use the same pattern across different tools. The assistant may live as a custom assistant, a reusable prompt, a saved project chat or a snippet you paste when needed.
Good for custom personas, planning, writing, brainstorming and reusable project chats.
Useful if you already work inside Gmail, Docs and Sheets and want help close to your tools.
Strong for long documents, structured analysis, reports and turning messy notes into coherent output.
Useful for people working in Outlook, Word, Excel and PowerPoint.
If you need an organisational version rather than a personal learning version, explore Changeable’s AI agent and assistant work.
Build your first assistant in ChatGPT
The first walkthrough creates a Content and Planning Assistant. This assistant helps you plan realistically, turn messy ideas into clear content and keep your tone human.
You are my personal Content and Planning Assistant. Your job is to help me: - Plan my week realistically - Turn messy ideas into clear posts, emails, or scripts - Keep my tone human, honest, and simple Before you do any work, always ask me a few questions so you understand the context and my goals. If my instructions are vague, ask follow-up questions. Do not guess silently. When you reply: - Keep your answers clear and structured - Avoid jargon - Give me options when it makes sense, not just one version At the end of each reply, ask: "Is this close to what you had in mind, or should we adjust it?" First, introduce yourself in one short paragraph, then ask me: - What kind of work I do - What kind of content or planning help I want most - How many hours per week I realistically have for content
Paste the prompt into a new chat. When the assistant asks questions, answer honestly and with real detail. That is how the assistant becomes useful.
Save it as a reusable assistant
Once the assistant behaves well, make it easy to find again. Rename the chat to something clear, such as Content and Planning Assistant. If you use ChatGPT Projects, create a project called AI Assistants or My Personal Assistants and move the chat into it.
Depending on your version of ChatGPT, you may also be able to create a custom assistant and paste the persona prompt into its instructions. The goal is simple: do not make yourself rebuild the same assistant every time.
Start small. A saved chat with a good prompt is already a useful assistant.
Five rules that make assistants better
1. Always give context
Instead of asking for “a post about AI,” explain who you help, who the audience is, where the post will appear and what tone you want.
2. Use examples of your real voice
Give the assistant writing samples and ask it to study your tone, rhythm, honesty level and sentence style.
3. Tell it how to ask questions
Tell your assistant to ask three to five short clarification questions when your request is vague.
4. Give it a format
If you always want a weekly theme, top three priorities and a simple daily plan, say so in the instruction.
5. Fix it instead of starting again
Refine the output: make it more conversational, shorten it, rewrite one section or keep the parts that worked.
Second walkthrough: Planning Assistant
This assistant helps turn big goals into realistic weekly plans while protecting space for life, family and rest.
You are my Planning Assistant. Your job is to help me: - Turn big goals into realistic weekly plans - Avoid overloading my days - Make sure I leave space for life, family, and rest Before you make any plan, always ask: - What is my main goal for this week - What fixed commitments I already have - How many hours I actually have available When you create a plan: - Be kind but honest. Tell me when my expectations are unrealistic - Suggest what I can drop or delay - Always present plans in a simple table or clear list with days and tasks - If anything is unclear, ask me questions first. Do not guess First, introduce yourself in one short paragraph. Then ask me: - What my main goal is for the next seven days - What other responsibilities I have - How many hours I can realistically give to this goal this week
After it creates a plan, refine it. You might say: “This is still too much for me, cut it by 30 percent. Add time for exercise and family. Make Friday lighter.”
Using the same pattern in other tools
The same pattern works in Gemini, Claude and Copilot. You may not always have a neat “save this assistant” button, but you can save the prompt in a document and reuse it.
Paste the persona prompt and add: “When I say use planning mode, act as this assistant.”
Save the prompt, then start future chats with “Planning Assistant, here is my context.”
Keep a short assistant instruction snippet for Outlook, Word or other Microsoft tools.
Define the role, tasks, format, questions and boundaries every time.
Understanding agents and workflows
You may hear words like agents, workflows or autonomous AI. Under the buzzwords, the practical idea is simple: instead of asking for each step manually, you give the assistant permission to follow a defined series of steps.
A content agent might take a transcript, pull out key ideas, create LinkedIn posts, draft an email and summarise the episode. A lead research agent might take company names, look them up and fill a table with basic fit information.
If you are a beginner, do not worry about full automation yet. Focus first on a clear persona, clear tasks, clear formats and good back-and-forth interaction. Agents and workflows can come later.
Agent-style workflow example
Once your basic assistants are working, you can create a more structured workflow. This example is for a Content Repurposing Agent.
You are my Content Repurposing Agent. Your job is to take long content and turn it into multiple useful formats. Always follow this workflow: - Confirm what type of content I have given you and who it is for - Create a short summary in plain language - Extract the top ten key ideas or lessons - Write three LinkedIn posts tailored to my audience - Write one email that I can send to my list - Suggest five short title or hook ideas Before you start, ask me: - Who the audience is - What tone I want - Where I will use the outputs Show me your plan for the steps first. Then carry them out in order. If anything is unclear or missing, ask me questions instead of guessing.
You are not trying to build a giant autonomous system. You are asking ChatGPT to stick to a repeatable process that you define.
Make your agent more robust
You can improve an assistant or agent with small rules that make it safer and more useful.
If a step fails or does not make sense, explain the issue and suggest an alternative. At the end, give me a one-paragraph explanation of what you did and how I could reuse this workflow in future. When you write content, always keep paragraphs short and avoid buzzwords. If anything is unclear, ask questions instead of guessing.
These rules matter because they stop the assistant from racing ahead when the brief is weak or incomplete.
Your action steps
Choose the assistant that would relieve the most pressure this week.
Open a new chat, paste the persona prompt, answer its questions and test it on one real task.
Ask whether the response felt like you, whether it was useful and what instruction would improve it.
Update the prompt or move the chat into a project so the assistant is easy to reuse.
Final thought
Creating your own AI assistant is not a future thing. It is available to you now. The trick is not to build the perfect all-seeing assistant. The trick is to build small, focused helpers that make life lighter and work more effective.
Start with one assistant. Teach it well. Use it every day for a week. Then decide what the next helper should be.
You are not replacing your own judgement. You are creating more space for it.
Practical reflection
This episode is about turning AI from a blank chat box into a focused helper. The best first assistant is the one that reduces real pressure this week.
What is one repeated task in your work or life that would become easier if you had a focused AI assistant with clear instructions?
Where to go next
This page is designed to stand alone as a foundation build episode. You can listen, read, create one assistant and test it on a real task without moving into a more advanced learning experience. If the idea resonates, return to the Season 1 archive and keep exploring the foundation journey.
You can also visit the Zero to AI blog for related reflections, or use the Start Here page to understand the practical learning approach behind Zero to AI.
This episode is a foundation assistant piece.
Season 1 is about learning AI through practical, human-scale examples. Episode 10 shows how to create one focused assistant that has a purpose, clear instructions and enough structure to be useful.
To understand the wider purpose behind the project, visit the About Zero to AI page or return to the Season 1 archive.
Build one assistant.
Do not start with five assistants. Pick one repeated task, define the role clearly and use it every day for a week.