1
Session Overview
Session 1 of your six-week programme. The focus was on foundations — what AI actually is, how to talk to it, and what tools you'll use. Sessions 2–6 are where the building starts.
| What we covered | Detail |
|---|---|
| What generative AI is (and isn't) | Large language models, hallucinations, limitations — and why relationships and IP still matter |
| Real client case studies | Three live examples showing what's possible: publishing, lead scoring, automated quoting |
| Your AI tool belt | The platforms you'll use across the six weeks — what they're for and how to access them |
| Prompting fundamentals | Role, context, instructions, format, examples — the building blocks of a good prompt |
| AI safety basics | Free vs paid models, sensitive data, prompt injection — what to watch for |
| Local language models | LM Studio — running AI privately on your own machine at no cost |
| Six-week programme overview | Where this is heading: ideation → MVP → deployment |
Where you're heading
By the end of six sessions you'll have a working minimum viable product (MVP) — a real app or website — built with AI. Today is about getting the foundations right so everything else is faster and easier.
2
Key Takeaways
The most important things from today. Keep these in mind as you start using AI tools this week.
Brief it like a new hire, not a search engine. AI is a genius with no common sense. Give it a role, context, clear instructions, and examples — the same way you'd onboard a talented assistant who knows nothing about your business yet.
Context first, then your question. AI reads in order. If you put your question first, it starts forming an answer before reading the important background. Role and context always come first.
One chat, one topic. Mixing unrelated tasks in the same conversation degrades quality. Start a fresh chat for each new task — AI gets confused in the same way a human would if you kept switching subjects mid-sentence.
Aim for 80% right, then refine. Don't expect perfection on the first response. Iterate: tell it what you liked, what to change, and what to cut. The first output is a draft, not a final answer.
Paid beats free for anything important. Free language models use your conversations to train themselves. Never put sensitive information, client data, or business-critical content into a free model.
Hallucinations are real — always verify. AI will sometimes state something confidently that is simply wrong, and it will justify the error if challenged. For anything factual or numerical, check the source.
Your relationships and IP are your moat. AI makes it easy for anyone to build an app or generate content. What protects you long-term is the trust you've built with customers and the unique knowledge baked into your process.
Local models are a powerful free option. LM Studio lets you run AI privately on your own machine — no subscription, no data leaving your device. Ideal for experimentation and sensitive work.
3
What AI Can & Can't Do
Understanding the limits is as important as knowing the possibilities.
✓ AI is strong at this
✓ Drafting, writing, and summarising content
✓ Researching and validating ideas quickly
✓ Generating and iterating on creative work
✓ Processing and structuring large datasets
✓ Coding, prototyping, and building MVPs
✓ Asking clarifying questions to shape your thinking
✗ AI struggles with this
✗ Replacing your judgment — human in the loop, always
✗ Guaranteeing accuracy — it can and does hallucinate confidently
✗ Maths and counting — use Excel formulas for calculations
✗ Building customer relationships — still a human skill
✗ Knowing your business without being told
✗ Truly original ideas — it remixes, it doesn't invent
4
Prompting Fundamentals
The quality of your output is almost entirely determined by the quality of your prompt. These are the five elements to include every time.
| Element | What it means | Example |
|---|---|---|
| Role | Tell it who to be — shapes vocabulary, depth, and perspective | "Act as an experienced business mentor with a background in digital startups." |
| Context | Share what it needs to know about your situation or goals | "I'm launching a freelance social media service targeting local restaurants in Essex." |
| Instructions | Be precise — what to do, what to cover, what to avoid | "Write a short pitch email introducing my service. Under 150 words. No jargon." |
| Format | Bullet list? Paragraph? Table? Specify — or it will guess | "Give me 3 bullet points and a one-sentence summary at the top." |
| Examples | Show it the style or output you want — especially for tone | "Here's an example of how I write: [paste sample]. Match this style." |
Here's a ready-to-use starter prompt structure you can copy and adapt:
Starter prompt structure ↓
You are a [role — e.g. experienced business mentor / marketing strategist].
Context: [Your situation — e.g. "I'm building a freelance [service] business targeting [audience] in [location]."]
Task: [Exactly what you want — e.g. "Help me write a 150-word pitch email introducing my service to potential clients."]
Format: [How you want the output — e.g. "Subject line, then 3 short paragraphs, then a call to action. No jargon."]
Tone: [e.g. "Warm and direct. Confident but not salesy."]
Ask me any clarifying questions before you start.
The iteration rule
Your first prompt is a draft, not a final instruction. When the output isn't quite right, don't start again — refine. Say: "Good — now make it shorter", "Change the tone to be warmer", or "Add a bullet point summary at the top." AI builds on the conversation.
✓ Do this
✓ Give it a role before asking your question
✓ Ask it to ask you clarifying questions first
✓ One chat per topic — start fresh each time
✓ Iterate and refine — treat the first output as a draft
✓ Include examples of the tone or format you want
✗ Avoid this
✗ Assuming it knows your business — always provide context
✗ Vague prompts like "make this better" — be specific
✗ Mixing unrelated tasks in the same conversation
✗ Expecting perfection on the first attempt
✗ Using AI for maths or counting — use a calculator
5
Your AI Tool Belt
These are the platforms you'll use across the six weeks. All have free access options. Get set up before Session 2.
Google Gemini
Free year for students
Huge context window (up to 1M tokens), excellent image creation with Imagen, connects naturally with Google tools. Mat's #2 pick — especially good for creative work. Free year available for students via Google's education programme.
gemini.google.com
Microsoft Copilot
Available via your uni account
Powered by GPT-4, built into your Microsoft account. Data stays within your organisational tenant — recommended for anything sensitive. Premium version gives access to Claude and other models. Included with your ARU student account.
copilot.microsoft.com
Google AI Studio
Main build tool for your MVP
Generous free tier. Builds full-stack applications directly in the browser and exports to GitHub for deployment. This is the platform you'll use to build your MVP in the later sessions of the programme.
aistudio.google.com
Claude (Anthropic)
Mat's personal favourite
Excellent at following detailed, complex prompts. Consistent, high-quality writing and nuanced responses. Also accessible via Copilot premium. If you're paying for one AI tool, Mat recommends this one. Free tier available.
claude.ai
Perplexity AI
Research & fact-finding
AI search engine that cites its sources. Great for validating business ideas, researching markets and competitors, and any task where you need traceable answers. Free tier available; student access may be available via ARU — ask Omkar.
perplexity.ai
LM Studio
Private, offline, free
Download open-source language models to run entirely on your own machine. No subscription, no data leaving your device, no usage limits. Slightly more setup required — a full walkthrough video will be in your shared resources folder.
lmstudio.ai
Netlify
Deploy your MVP — Session 6
Free hosting platform for your finished application. Used in the final session to make your MVP live on the internet. Free tier is expected to cover basic deployment — no cost required. You won't need this until the final sessions.
netlify.com
NotebookLM (Google)
Research & document analysis
Upload PDFs, articles, or notes and ask questions across all of them. Great for synthesising research or understanding complex documents quickly. Worth exploring independently alongside the main tool belt.
notebooklm.google.com
All tool links in one place
The full tool belt is also available at link.aiguys.ltd/ARU — bookmark it for quick access throughout the six weeks.
6
What's Actually Possible — Three Client Examples
Real projects built by The AI Guys. These are here to show what's achievable once you know how to use these tools — and to spark ideas for your own MVP.
Inspired By — Publishing Portal
60% efficiency gain
The problem
Manually reviewing and giving detailed feedback on each client's book chapters was the biggest drain on the team's time. One reviewer, many clients, and a growing waiting list.
What was built
A custom AI portal trained on the company's own review framework. Clients upload chapters; the AI gives structured, specific feedback based on the firm's unique IP and editorial standards.
The result
60% increase in efficiency. Capacity to take on 60% more clients. The portal is now being developed as a resaleable SaaS product — a new revenue stream born from the build.
University of Suffolk — AI Business Development System
Thousands of leads/week
The problem
Manually researching and scoring potential business partners and placement employers from a small, manually maintained list — hours of work per week for a limited output.
What was built
An AI system that finds companies matching a set of defined criteria, researches each one automatically, and scores them based on fit — without any manual input required.
The result
Moved from processing 50–60 leads per week manually to thousands processed automatically. Countless hours saved. A custom portal is now in development to extend the system further.
Process Instrument Solutions — Automated Quoting
3 hours → ~1 minute
The problem
Quote requests of 100–150 items arrived by email. Each required opening multiple PDFs, finding specs, locating prices, and manually building the full quote. One quote could take three hours or more.
What was built
An AI quoting system using RAG (retrieval-augmented generation). Staff paste the quote request; the AI reads it, searches the product database, and produces the complete quote automatically.
The result
Quote turnaround reduced from up to 3 hours to approximately 1 minute per quote. Direct email integration is in progress, which will remove the manual paste step entirely.
The pattern to notice
In each case, the AI didn't replace the business — it amplified it. The firm's knowledge, relationships, and IP stayed central. The AI handled the repetitive, time-consuming process built around it. That's the model worth following as you design your MVP.
7
AI Safety — What to Watch For
Most of the risks with AI come from how the tools are used, not the tools themselves. Keep these in mind from day one.
Free models use your conversations for training. If you're using the free tier of ChatGPT, Gemini, or similar, your prompts and responses may be used to train the model further. Never enter sensitive client data, personal information, or confidential business details into a free model.
Always keep a human in the loop. AI can confidently produce plausible-sounding content that is factually wrong. For anything that matters — figures, facts, proposals going to clients — verify before using. Don't let it make decisions for you.
Be cautious with agentic AI. Agent mode in ChatGPT, Claude, or Copilot can take actions on your behalf — browsing, sending emails, executing tasks. Powerful, but risky. Don't give agents access to your email account, banking, or sensitive systems without understanding their limits.
Understand prompt injection. If an AI agent is browsing the web on your behalf, attackers can hide instructions on web pages (in invisible text) that hijack the AI. Be careful about what agentic tools have access to, and don't let them browse without guardrails.
Paid accounts offer better privacy. With a paid subscription, your conversations are generally not used for model training. For business use, this is worth the investment — especially when working with client or commercially sensitive information.
8
Prompting 101 — Technique Quick Reference
The full Prompting 101 Cheat Sheet is in your shared resources folder with complete templates and worked examples. This is your quick-reference guide to choosing the right technique for the task.
| Technique | When to use it | The one-line version |
|---|---|---|
| Instruction-Style | General writing, rewrites, summaries, everyday tasks | Role + Task + Format. Simple and effective for most things. |
| Context + Question | When your situation matters — tailored advice, document Q&A | Background first, then your question. Reduces generic responses. |
| Chain-of-Thought | Complex problems, multi-step decisions, strategy | Ask it to think step by step before giving the final answer. |
| Self-Critique / Self-Refine | High-stakes writing, proposals, anything needing accuracy | Ask it to answer, critique itself, then give an improved version. |
| Verbalised Sampling | Brainstorming, ideation, creative options | Ask for 3–4 options with confidence levels. Opens up more creative responses. |
| Prompt Repetition | Classification, extraction, simple factual queries | Repeat your key instruction at the end in slightly different words. |
| Anti-Prompt Approach | When you're tired of generic, wordy, corporate-sounding responses | Tell it what NOT to do. No fluff. No jargon. Single best answer. Go. |
| CRAFT Framework | Marketing copy, audience-specific content, reports | Context, Role, Action, Format, Target Audience. |
| CARE Framework | Summaries, transformations, structured outputs | Context, Ask, Rules, Examples. Great when examples really help. |
| RTF Framework | Quick daily tasks, coding, documentation | Role, Task, Format. Fast and effective for routine requests. |
Try the Anti-Prompt quick template — paste this before any task to immediately improve output quality:
Anti-Prompt quick template ↓
Role: [Your role — e.g. startup founder / freelance designer / marketing manager].
Experience: [e.g. early-stage, 2 years in the industry].
Tone: Direct.
Task: [Your question or request here]
Rules: No options. No fluff. No introductions. No "certainly!" or "great question!".
Single best answer only. Go.
Where to find the full guide
The Prompting 101 Cheat Sheet is in your shared resources folder. It includes full templates, worked examples, and the complete Anti-Prompt master template. Use it alongside this crib sheet as your go-to reference throughout the six weeks.
9
Try This Today
Don't wait until Session 2. These five tasks will give you a feel for the tools and prompting approach — and set you up to hit the ground running on ideation next week. Click each one to mark it done.
✓
Test the same prompt across three tools
Take one question or task relevant to your business idea. Run it through Copilot, Gemini, and Claude (or whichever three you have access to). Compare the quality, style, and usefulness of each response. Note which you preferred and why.
✓
Set up your free Gemini year
Sign into gemini.google.com with your student Google account. The free year of Gemini Advanced is available via Google's education programme — follow the link in your shared resources folder or ask Omkar for the direct link.
✓
Read through the Prompting 101 Cheat Sheet
It's in your shared resources folder. You don't need to memorise it — just read through it once so you know what's there. The techniques will make much more sense once you've had a go at prompting.
✓
Build a context block for your business idea
Write a short paragraph describing your idea, your target customer, and your goals. This becomes your reusable context block — paste it at the start of any prompt so the AI always knows who you are and what you're building.
✓
Download LM Studio and have a look around
Visit lmstudio.ai and download the app. A full walkthrough video will be in the resources folder. It's free, runs entirely on your machine, and has no usage limits.
10
What's Coming in Session 2
Session 2 is where the building begins. Here's what to expect — and how to prepare.
| Focus area | What you'll do |
|---|---|
| Ideation with AI | Use Copilot (or your preferred model) to develop and pressure-test your business idea through structured back-and-forth conversation |
| Ideal Customer Profile (ICP) | Use AI to ask you the right questions and help you identify exactly who you're building for — before you start building |
| Brand asset creation | Generate your logo, brand colours, and tone of voice using AI image and writing tools |
| Hands-on throughout | Sessions 2–6 are all practical. Come with your idea (even a rough one) and be ready to build |
Come prepared
Think about what you'd like to build before Session 2. It doesn't need to be fully formed — a rough idea is enough. We'll use AI to refine it together. If you're completely stuck on an idea, that's fine too — we'll use AI to help you find one.
Shared Resources Folder
Session slides, this crib sheet, the Prompting 101 Cheat Sheet, tool links, and future session resources — all in one place. Ask Omkar for the link if you don't have it yet.
Questions Between Sessions
Got a question or want a walkthrough video of a specific tool? Let Mat or Omkar know and it'll go straight into the resources folder.