The Matts Workshop

Digital Productivity & AI in the Workplace

Getting Started

AI Copilot Exercises

Click an exercise to reveal the prompt. Copy and paste it into Copilot.
Important: Always click "New Chat" between tasks.

Exercise 1: Explain it like I am 5 (ELI5)
Application: Quickly understand complex topics or prepare simple explanations.
"Explain [TOPIC] to me like I am 5 (ELI5)"

Replace [TOPIC] with: "cloud computing", "DNA", "quantum physics", "offside rule in football" etc.

Exercise 2: Roleplay for Difficult Conversations
Application: Rehearse performance reviews or conflict resolution in a safe space.
"Roleplay: You're Mrs. Grudge, my grumpy neighbour knocking at 2 a.m. to complain about my noisy cats (Knox, Missy, Phoebe). I'm a witty cat owner trying to calm you down and win you over. Stay in character and wait for my replies."
Exercise 3: Devil's Advocate (Blind Spots)
Application: Stress-test your challenges, new ideas, or decisions before finalizing them.
"Act as a constructive devil's advocate. First identify potential weaknesses or blind spots, then suggest alternative perspectives. Be specific and challenging but constructive. My scenario is: [scenario here]"

Add in your own scenario: "I want to move from sales to leadership role", "dealing with a difficult client", provide some useful context as well.

Exercise 4: Professional Persona (Role Prompting)
Application: Tailor communication tone for specific audiences.
"For a script: You are a medical professional. Reply to the statement 'My haemorrhoids are bothering me'."
"For a script: You are a comedian. Reply to the statement 'My haemorrhoids are bothering me'."
Exercise 5: Format & Structure
Application: Force the AI to output exactly what you need (Table, SOP, Creative Text) in a single request.
"You run 'Knights of the Round Table Pizza' - a medieval-themed pizza delivery service. Please generate the following three items:

1. Write a catchy jingle for your horse-drawn pizza delivery adverts.
2. Draft an SOP for training delivery knights on horseback pizza transport and armour maintenance.
3. Create an Excel table with three columns: Medieval pizza names, ingredients, and delivery times to different kingdoms."

Exercise 6: Self-Evaluation (Mark Your Homework)
Application: Get AI to critique its own work to improve quality before you review it.
"I need a name for a new pensioner dating app, provide five examples please. For each name justify its selection and ensure its not condescending or offensive to the target market, though don't be afraid to be humorous. Score each example on a 1-10 scale for Memorability, Distinctiveness and Global potential. Provide a reason for the score."
Exercise 7: Prompt Chaining
Application: Combine Role, Tone, Constraints, and Format into one request.
"Task: Suggest appealing fizzy drink names. Role: Young consumers (15-25). Tone: Humorous, but not childish. Constraint: Avoid existing UK product names. Format: Provide 10 names in a 3-column table: Name, Rating (1-10), Reasoning Examples: CitrusBurst, AquaFizz, SolarPop."
Exercise 8 – Create My First Agent
What are Agents? They are reusable Copilot assistants with a specific focus – saved prompts with personality that you can share with your team.
  1. Open Copilot – go to m365copilot.com and click "Create agent" in the left-hand menu.
  2. Enter Description then submit – In the "Describe" tab, type the following in the box at the bottom and hit the submit arrow:
    "The user will provide a word and you must generate 3 possible crossword style clues for that word"
  3. Click Create You might get prompted to give the Agent a name, reply to that then click the Create button.
  4. Click Go to Agent – Wait for the success message, then click the blue 'Go to Agent' button.
  5. Test e.g. Book – Type a single word like "Book" into the chat to see if it generates 3 crossword clues.
Exercise 9 – Build a New Agent: Sentence Categorizer
New Agent Task: In this exercise, you will build a new agent designed for a completely different task: analyzing text data and generating visual word clouds using Python.
  1. Open Copilot – Go to m365copilot.com and click "Create agent" in the left-hand menu to start a fresh agent.
  2. Describe the Agent – Ensure you are on the Describe tab. Copy and paste the description below into the chat box and hit the submit arrow:

    πŸ“‹ Copy this Prompt:
    "The user will provide sentences (via attached .xlsx or pasted below). For each sentence, assign only one broad, high-level parent category as a SINGLE word (e.g., generalize β€˜Spreadsheet’ to Work, β€˜Tacos’ to Food). Use CONSISTANT category names and avoid splitting similar concepts into separate categories. REVIEW YOUR CATEGORIES FOR CONSISTANCY ONCE YOU HAVE BEEN THROUGH THE FULL LIST.

    DO ALL OF THE FOLLOWING 3 THINGS AUTOMATICALLY:
    1, Output a table with columns: Sentence, Category.
    2, Calculate topic frequency by counting how many sentences fall under each category.
    3, Write and Execute Python code to generate and DISPLAY on screen a word cloud where each category’s size is proportional to its topic frequency."
  3. Configure Settings – Switch to the Configure tab (next to Describe). Name your agent "Sentence Categorizer".
  4. Enable Code Generation – Scroll down to the "Capabilities" section. Toggle the slider for "Create documents, charts, and code" to the On position (it will turn blue).
  5. Create Agent – Click the Create button and wait for the success message.
  6. Test it – Click "Go to agent". Use the sample data below (either upload the file or copy the text) to see it generate a category table and a word cloud image.

πŸ“‚ Sample Data for Testing

You can either download the Excel file called sentences.xlsx to upload to your agent, or copy the text below directly into the Agent's chat.

Or copy these sentences:
Meta-Prompting: Let AI Write the Prompt
Application: Leverage the AI's knowledge to design expert prompts for tasks in specific domains e.g. 'Triage Helpdesk Emails in IT'.
"You are an expert AI prompt designer. Please design the most effective prompt to: [INSERT SPECIFIC TASK AND DOMAIN]. Consider what key elements, criteria, or details matter, what output format is most actionable, and the reasoning steps that are essential."
Reverse Engineering (Creating Prompts from Gold Standard Outputs)

Concept: "Meta-Prompting" or Reverse Engineering. Instead of guessing how to prompt, feed the AI the desired output (or a stylistic reference) and ask it to write the prompt for you.


1. Meeting Notes Template

Application: Use this when you have a "Gold Standard" document and you want all future documents to follow that exact structure.
"Analyze the meeting notes below. Create a single, reusable prompt that instructs an AI to generate notes using this exact format, style, and structure from a raw transcript.

[PASTE GOLD STANDARD NOTES HERE]"


2. Corporate Tone Persona

Application: Essential for marketing. Use this to extract a specific "Voice" from existing copy so the AI can mimic it later.
"Analyze the tone, vocabulary, and sentence structure of the text below. Write a 'Persona Prompt' defining these characteristics so an AI can consistently write new articles in this exact style.

[PASTE BRAND TEXT HERE]"


3. Academic Abstract Summarizer

Application: Use this for technical writing where constraints (word count, objectivity) are strict and you want to summarize new content in that voice.
"Deconstruct the academic abstract below. Identify the implied word count constraints, formal tone, and structural elements. Write a reusable prompt designed to generate similar abstracts from new content.

[PASTE ABSTRACT HERE]"


4. Image-to-Text Style Transfer

Application: Use this to generate a detailed text description of an image, which you can then use to generate new images in that style.
"Act as a professional photographer. Analyze the attached image. Write a detailed text-to-image prompt capturing the specific lighting, mood, camera angle, lens type, and environmental details to recreate this style.

[UPLOAD IMAGE]"


5. Website Style & Format

Application: Use this to extract design schemas from screenshots or HTML code to produce new sites with a similar look and feel.
"Analyze the visual design, color palette, typography, and layout of the website below. Write a design system prompt that instructs an AI to apply this exact look and feel to new code.

[PASTE URL, HTML, OR UPLOAD SCREENSHOT]"
AI Usage Case: Meeting Notes
"Please summarize the following meeting transcript. Focus on the key points, decisions made, action items and any important deadlines. Ensure the summary is clear and organised. The summary should be no longer than 200 words: [INSERT TRANSCRIPT HERE]"
AI Usage Case: Excel Data Analysis
"Summarise this Excel file: PROVIDE: 1. OVERVIEW (rows/columns, data type, date range) 2. KEY STATISTICS (totals, averages, min/max) 3. TOP INSIGHTS (3-5 interesting findings) Present in clear business language."
AI Usage Case: Text Categorisation
"Analyze the [Name of Feedback column] column. Add a new column named 'Category'. Classify each row into exactly one of these specific categories: 'Compensation', 'Management', 'Work Environment', 'IT/Tools', or 'Work-Life Balance'. If the feedback doesn't fit these, label it 'Other'."
Responsible AI (Best Practices)
The Human in the Loop: AI is a tool to enhance your work, not replace your judgment. Always keep these three pillars in mind.
🀝
Human-Centric Approach
  • It's your assistant (augment over replace).
  • Protect critical skills: use AI to enhance them.
  • Target repetitive manual tasks first.
βœ…
Practical Checks
  • Always double-check the output.
  • Be Transparent: "I used AI for this."
  • Protect your data (use IT approved AI tools).
βš–οΈ
Responsibility & Ethics
  • Watch out for AI's blind spots (bias/hallucinations).
  • Own the results: AI doesn't take the blame.
  • Align your usage of AI with your values.
It's Your Teammate (The Conversation)
The Reality: You rarely get a perfect draft from a human colleague on the first tryβ€”don't expect it from AI. The magic happens in the follow-up.
Comparison: The Vending Machine mindset vs The Teammate mindset
πŸš€ Use the "Teammate" Boilerplate Prompt

As illustrated on the right side above, stop asking for instant, giant deliverables. Instead, frame the interaction as a guided collaboration using this structure:

"I want you to help me [INSERT BIG TASK HERE], but don't try to solve it all at once. Instead, act as a helpful guide..."
🀝
The "Teammate" Mindset

Stop treating Copilot like a vending machine (input coin -> get product). Treat it like a clever intern.

If the output isn't right, don't rewrite the prompt from scratch. Just talk to it. Tell it what it did wrong and guide it to the answer.

Try these follow-up phrases:
"Actually, can you make it more formal?"
"That's good, but what if we tried a friendlier approach?"
"Can you expand on that second point? It's a bit too brief."
Practical Prompting (The 1-2-3 Method)
The Golden Rule: Better inputs equal better outputs. Follow this 3-step flow for every prompt to get the best results.
1.
Give Context

Provide relevant background information (Role, Data, Reports).

πŸ‘€ πŸ“Š 🌐 πŸ“„
2.
Make Request

Be specific about exactly what you want the AI to do.

"Review my email for clarity..."
3.
Explain Purpose

Share why you need this (The "So That").

"...to maximize the chances of a quick response."
AI Hacks - MegaDoc (Context Booster/External memory)
Concept: Stop typing the same background info. Create a simple Word doc with your org's specific context to copy-paste into prompts.

Without help, AI will guess what your acronyms mean and write in a generic, robotic tone. A "MegaDoc" solves this by giving the AI immediate situational awareness.

πŸ”€
Acronyms

Define your internal language (e.g. "DCG = Division of Cancer & Genetics")

βœ’οΈ
Style Guide

Set the tone: "Formal but warm," "Use UK English," "Use bullet points"

πŸ“‹
Key Facts

Boilerplate info, key personnel names, and standard operating procedures

πŸš€
Project Overview

Goals, status, and blockers. Give the AI immediate situational awareness

Create your own today:

Open a blank document and list your top 10 acronyms and one Project Snapshot. The next time you prompt, paste it in to see the difference immediately.

"Here is some context from my MegaDoc: [PASTE YOUR ACRONYMS/STYLE GUIDE HERE] Based on this context, draft an update email to the team about [Your Topic]..."
Private AI (Running Models Locally)
You Are In Charge: Running AI locally using on-device storage offers a secure environment, total privacy, zero fees, and works without Wi-Fi.
πŸ›‘οΈ
Total Privacy & Control

Data Sovereignty: No data leaves your device.

  • Safe for sensitive/classified info.
  • Full IP protection & ownership.
  • No third-party platform risks.
⚑
Cost & Efficiency

Operational Benefits: Lower overheads.

  • Zero Fees: Open source models
  • Offline Capable
  • Lower energy consumption.
πŸŽ“
Private AI Workshop

We run a 2-hour practical session to get you started with Private AI on your laptop.

Contact us for details β†’
Your AI Road Map

No one knows everything about AI, but together we can figure it out faster.

This is the AI Community Approach.

πŸ“„ Read our Paper on the AI Community Approach β†—
🧠
Mindset

Be curious. Ask: "Could this be useful?"

πŸ’¬
Conversations

Shared learning, wins, and fails.

πŸ§ͺ
Experiments

Learn by doing. Build expertise.

🀝
Community

Turn scattered experiments into shared learning.

Contact Information

Questions? Feel free to contact us:

Dr. Matt Mort: mortm@cardiff.ac.uk
Mr. Matt Hayden: haydenmj1@cardiff.ac.uk