Older software could store this content, but it could not search it, summarize it, or use it well.
→
THE TOOL
An LLM finds structure in messy information.
Conversation is only the interface. The real job is to make sense of information. Later today, you will use this on your own research.
6 · HOW LLMs WORK
One token at a time
INTERACTIVE · CLICK ALONG
① TEXT BECOMES TOKENS
② THEN: PREDICT THE NEXT ONE
An LLM writes its answer one
The model did not think. It predicted the most likely next token, again and again. This is why AI stops feeling like magic.
7 · RAG
How an LLM reads your document
LIVE DEMO
RAG means Retrieval-Augmented Generation. Every time you paste research into ChatGPT, this process runs.
Your documenta file the model has never seen
→
Split into chunkssmall searchable pieces
→
Retrievefind the pieces relevant to your question
→
Generateanswer built from those pieces
Demo: upload a short PDF. Ask a question that only that document can answer. The reply should include a detail that was never in the model's training. This proves that the model read the file instead of guessing.
8 · MCP
From describing to doing
On its own, an LLM can only produce text.
MCP (Model Context Protocol) connects it to your tools, so it can take action there instead of only telling you what to do.
“We started with what is intelligence, and we end with a model that can read your documents and act inside other systems. That is exactly what you'll use for the rest of today.”
HOUR 1 ✓ · THE VOCABULARY IS SET
Five terms, now clear
AI
The broad term: any machine that does work that normally needs human intelligence.
LLM
The language part of AI: ChatGPT, Copilot, and Claude.
Token
The small pieces that text breaks into. The model predicts them one at a time.
RAG
How an LLM reads and searches your document before answering.
MCP
How an LLM connects to other tools and takes action, not only text.
2
HOUR 2
AI Mindset Tool → Co-Worker
Hour 1 gave you the concepts. This hour turns them into a working habit: when to do a task yourself, when to work with AI, and when to give the full task to AI.
OPENING QUESTION
?
Think of the last time you asked AI for something and the result was disappointing. Raise your hand.
Now think back to what you actually typed. Did you give it enough context to know what you needed, or did you expect it to guess?
1 · REALISTIC LIMITS FIRST
Strengths vs Weaknesses
LIVE DEMO
STRENGTHS
✓
Drafting: first versions of any text
✓
Summarizing: long input, short output
✓
Structuring: turning raw notes into outlines, tables, and plans
WEAKNESSES
✗
Decisions that depend on context it does not have
✗
Office politics and unstated priorities
✗
Incomplete information: it cannot know what you did not tell it
Demo: ask ChatGPT Should I approve this budget request? with no details. It will either ask you for more information, or give a generic answer that you cannot use. This shows the limit, live.
2 · THREE WAYS TO WORK WITH AI
Chat · Co-Worker · Assistant
INTERACTIVE
Chat
One question, one answer. Similar to a search engine.
Co-Worker
A conversation with many turns. AI keeps the context and shapes the work with you.
Assistant
AI repeats a defined task with little supervision. Named today, not demoed.
you · one prompt onlywrite an email about a delayed project
ai · first answer, used as it isDear Team, I am writing to inform you that the project has unfortunately been delayed. We apologize for any inconvenience this may cause. We are working diligently to resolve the issues and will provide updates as soon as possible…
Generic and formal. No names, no dates, no reason. Difficult to use.
turn 1write an email about a delayed project
turn 2make it shorter
turn 3the tone is too formal, fix that
final resultHi team, quick update: Phase 2 moves to next Thursday. We found an integration issue and we want to fix it properly. You do not need to do anything this week. I will share the new timeline tomorrow.
Three turns, same tool. You shaped the result, and AI kept the context. This is Co-Worker mode. Most people never discover it.
3 · THE DELEGATION FILTER
Yourself · AI-supported · AI-delegated
CLICK A CARD TO REVEAL
Every task goes into one of three groups.
Do it yourself
With AI's help
Give it to AI, then review
Write a weekly status update?AI writes the draft from your bullet points. You add the judgment and the names. This takes minutes, not an hour.
Research a competitor?Give the full search to AI, then review the results. You spend your time checking, not collecting.
Decide a strategic recommendation?This depends on politics, priorities, and context that AI does not have. AI can prepare a summary, but the decision is yours.
30 MINUTES
Break
30:00
We return in 30 minutes. Next: what to delegate, and which tool to use for it.
3
HOUR 3 · AFTER THE BREAK
AI Use Cases What to delegate
From mindset to specific tasks: writing, research, and visuals, with the right tool for each one.
OPENING QUESTION
?
Have you ever asked ChatGPT to build a presentation? Or to research a topic and give you sources?
How did the result look?
ChatGPT is built to do a bit of everything. It is not built to do any one of those things best.
1 · WRITING & COMMUNICATION
The most reliable use of AI
LIVE DEMO
The input is text and the output is text, so nothing is lost. We start here because most people already use AI this way.
PASTE THIS ROUGH PARAGRAPH…
hi so basically the thing is the vendor didnt send the report again and we cant close the monthly numbers, i think we need to maybe escalate it or something because finance keeps asking me and idk what to tell them anymore
THEN ASK, ONE BY ONE:
Rewrite this message for clarity and a professional tone.
Now shorten the same message into a two-line Slack update.
Show both outputs side by side: the email version and the Slack version.
The same rough input becomes two clean formats. This is one skill you will reuse daily: drafting, editing, and summarizing.
2 · RESEARCH
Research needs sources
LIVE DEMO · PERPLEXITY
Research means finding and summarizing information from outside sources, not from a document you already have. (Reading your own document was RAG, this morning.)
Perplexity is built for exactly this: it searches the web and shows its sources. ChatGPT can do a lighter version of the same thing.
ASK PERPLEXITY, LIVE:
What are the most common reasons customers stop using a subscription service? Give me sourced findings I can compare.
Point at the citations as they appear: the findings have sources, and you can compare them. This is not one unverified answer. Most people stop at their first search result. This method is better.
3 · VISUAL & SLIDE CREATION
Presentable, without a designer
LIVE DEMO · GAMMA + IMAGES
GAMMA · OUTLINE → DECK
Pitch: reducing customer churn
The problem: 30% of cancellations happen in month one
The fix: guided onboarding + an early warning signal
The ask: a 3-month pilot with the support team
Paste these four lines into Gamma. It builds a formatted deck in less than a minute.
IMAGE GENERATION · TEXT → VISUAL
A clean, minimal illustration of a customer journey from confused to confident, corporate style, blue and white palette
One short prompt creates a custom image that no stock library has.
You will see these same tools again in Hour 4, when we build the pitch.
4 · MATCH THE TOOL TO THE TASK
Match the tool to the task
CLICK EACH ROW TO REVEAL
General writing & drafting?ChatGPT · Copilot
Research with sources?Perplexity
Longer, structured writing & analysis?Claude
Slide decks?Gamma
Custom visuals?Image tools
This map replaces the question "which AI tool do I use?" with a clear default. You will use it for the rest of today and in your own work.
4
HOUR 4 · FINAL HOUR
Working effectively, then building a pitch live
A short block on getting better output, then one continuous demo: research packet → insight → story → slides → rehearsal. The hour is compressed on purpose, so it moves fast.
OPENING QUESTION
?
Think of a time when the AI result was wrong or weak. What did you actually type?
And the last time you built a pitch without AI, where did the time actually go?
Structure, writing, visuals, rehearsal: the time goes everywhere at once. So the demo ahead covers all four stages, not only one.
1 · WRITING BETTER PROMPTS
A strong prompt states four things
TURN EACH PART ON
Rolewho the AI should act as
Contextwhat it needs to know
Goalwhat result you want
Formathow the result should look
YOUR PROMPT
write about our new product launchYou are a marketing writer.Our new product has just launched; the audience is existing customers.Write a post announcing the launch.Make it a 100-word LinkedIn post, upbeat but professional.
output quality: vague in, vague out
Then prove it live: run the weak prompt and the full four-part prompt in ChatGPT, side by side. The difference is clear without explanation.
2 · REUSABLE INSTRUCTIONS
Stop retyping your best prompt
LIVE DEMO
ChatGPT and Claude let you save instructions once. These are called custom instructionsskillsmemory. They apply automatically to every future request.
Write the four-part prompt once and save it. It then works for every future request.
This turns a one-time trick into a repeatable habit.
Live demo: one saved instruction, and a request that benefits from it automatically.
3 · BUILDING A PITCH · RESEARCH → INSIGHT → STORY
From research to story: three prompts, back to back
LIVE DEMO
THE RUNNING EXAMPLE: PASTE THIS FICTIONAL RESEARCH PACKET
Research notes: "StreamFit", a fitness subscription app
• 30% of cancellations happen in the first month
• 61% of churned users say they "didn't know where to start"
• Users who finish 3 workouts in week one renew at 2× the rate
• Support tickets peak on days 2–5, mostly "how do I…" questions
• Competitors with guided onboarding report 18% lower first-month churn
THEN CHAIN: EACH OUTPUT FEEDS THE NEXT PROMPT
1
What is the single core insight in these notes? One sentence.
2
Who is the likely audience for this pitch, and what would they object to?
3
Using that insight and that audience, build a three-part story arc: problem, solution, impact.
Keep the objection from prompt 2 on screen. It returns in the final step.
4 · BUILDING A PITCH · STRUCTURE, SLIDES, VISUALS
The story becomes something you can see
LIVE DEMO · CHATGPT → GAMMA
OutlineAsk ChatGPT for a slide-by-slide outline from the story arc.
→
SlidesPaste the outline into Gamma. It builds a real deck in a minute.
→
One visualSupport the strongest claim (the 2× renewal number) with a simple chart.
Turn this story arc into a slide-by-slide outline for a 5-slide pitch. For the two most important slides, write the exact headline and one supporting line.
This reuses skills from earlier today: prompting from this hour, and Gamma from Hour 3.
5 · BUILDING A PITCH · STRENGTHEN & REHEARSE
AI does not only build the pitch. It also makes the pitch stronger.
LIVE DEMO
FIND THE WEAKEST POINT
List the 3 hardest questions this pitch's audience would ask.
The callback: before you run this, point to the objection from step 3. The hardest questions should include that same objection, not a new unrelated list. If it is missing, ask AI why.
REHEARSE THE DELIVERY
Give delivery notes on the opening and closing lines: pacing, tone, and where to pause.
Close the day here. The finished pitch elements are on screen: insight, story, slides, visual, hard questions, and delivery notes. This is the promise of the day: AI that strengthens a pitch, not only builds it.
×
END OF DAY
This morning, "AI" was a buzzword. Now it's a colleague.
✓ Five terms made clear: AI · LLM · Token · RAG · MCP✓ A delegation filter for any task✓ A tool map: the right tool for each job✓ Four-part prompts✓ A pitch built and tested, live
The next step is tomorrow morning: choose one real task, apply the filter, and let AI write the first draft.
شكراً · thank you · the go-to place for AI & tech education
00:00
✕
Presenter shortcuts
→Space
Next (reveals build steps first)
←
Back
N
Presenter notes: your teaching script per slide
F
Fullscreen
T
Elapsed timer on/off
Home / End
First / last slide
?
This help
Interactive slides: click the highlighted elements on screen. They are part of the session.
×