A session preview

Today's Session

• 01 · AI Basics — what it can (and can't) do · 30 min
• 02 · Build Something Real — your idea, live · 105 min
• 03 · Thinking at Scale — beyond today · 30 min
• 04 · Wrap-up & next steps · 10 min

Right Now

Before we start — keep this in mind all day. What's one task you'd love AI to take off your plate? Hold onto your answer — we'll build toward it, and check back at the end of the day.

You Already Told Us · Survey Q1

How you're already using AI today: 30 of you responded — 21 already use AI in some form, 9 haven't started. That's exactly who today is for.

You Already Told Us · Survey Q2

What you want to walk away with: of the 30 who shared, the #1 ask was making everyday work faster and more practical. 5 flagged data safety and responsible use — we cover it.

By The End Of Today

• A working answer to what AI can and cannot do — and why it hallucinates
• A real problem statement and PRD for something that slows down your work
• A working tool you built yourself, live on the internet with your own link
• A way of thinking about AI that goes beyond one chat window

What Is AI, Really

AI is not magic. It reads input, applies logic, and produces output. A language model reads text, finds patterns, and generates a response — it does not think, it predicts what the most useful next output should be.

Two Types Of Tasks

Things AI does reliably, and things it does not — knowing the difference is the whole game.

• Does reliably: summarise long documents, draft emails/letters/reports, reformat messy notes, brainstorm in volume, translate and simplify language
• Doesn't do reliably: judge ambiguous human situations, verify facts it wasn't given, replace professional judgment, guarantee accuracy in specialised cases, know when it doesn't know

Why Hallucinations Happen

An AI model is always guessing the next most likely word — even when it doesn't know the answer. Guessing a plausible answer is rewarded more, during training, than admitting uncertainty.

What This Means For You

Treat AI like a fast, confident junior assistant — not a source of truth.

• Check anything specific: names, numbers, dates, quotes, policies
• Give it the source document instead of asking it to "recall" facts
• Treat confident-sounding answers with the same scrutiny as unsure ones

Output Quality

Garbage in, garbage out. Output quality depends entirely on the quality of your instruction — not on how "smart" the AI is.

A Good Prompt Has 4 Parts

Role. Task. Context. Output format.

• Role — tell the AI what it is
• Task — one clear action
• Context — what it needs to know
• Output format — exactly what you want back

The Coursera Take

“Is it perfect? No. Is it really, really valuable? Yes.”Jeff Maggioncalda, CEO · Coursera

The Path Today

From idea to live website — five steps, done live, the same path anyone can walk again tomorrow: Problem → PRD → Build → Deploy. No code typed by hand — you describe what you want, in plain English.

Step 1 — Define Your Problem

What's one task in your work that feels repetitive, slow, or tedious? It doesn't need to be big — small, real, and specific beats big and vague.

Step 2 — Turn It Into A Problem Statement

One sentence: who has the problem, what it is, and why it matters. Template: "[Who] struggles with [what], which causes [impact]."

Step 3 — What Is A PRD?

A Product Requirements Document — a short brief that tells AI exactly what to build. It removes guesswork: goal, who it's for, what it does, what it looks like.

A Simple PRD Has 5 Parts

• Goal — what problem this solves, in one line
• Users — who will actually use this
• Features — the 2–3 things it must do
• Look & feel — simple, colours, one page or several
• Out of scope — what it does NOT need to do yet

Step 4 — PRD To A Working HTML File

Paste your PRD into ChatGPT, Gemini, or Claude, and ask it to build a single, self-contained HTML file — one file, no setup.

Step 5 — Deploy To Netlify

No code. Drag your HTML file onto app.netlify.com/drop and get a live, shareable link in seconds — no account needed to try it.

Bonus · The Pro Path

How Haojun does it day to day, one level up: Claude Code → GitHub → Vercel → Custom Domain. GitHub gives full version history and code ownership; Vercel auto-deploys every change with free SSL; a custom domain looks professional instead of a "netlify.app" link.

Beyond The Chatbot

AI is more than one chat box — different jobs, different tools, most needing nothing more than a description of what you want.

• Make slides & decks — Gamma, NotebookLM
• Design posters & graphics — Canva Magic Studio
• Edit video & audio — Descript
• Do multi-step work on your files — Claude Cowork

Recap — What You Built Today

• A clear problem statement, in your own words
• A PRD that tells any AI tool exactly what to build
• A working HTML tool, built with AI, not by hand-coding
• A live, shareable link — a real website, not a mockup

Two Levels Of AI Thinking

Day-to-day: small wins that compound. Organisational: how you deploy AI at scale.

Rolling AI Out At Scale

Three phases, in order — don't skip ahead.

• One team, one real problem, controlled
• Fix what breaks, document what changed
• Widen access, track adoption — not just usage

People Don't Resist Technology

They resist losing control, looking incompetent, or being replaced. Bring people along before the tool goes live — let them find the first bug, that's how champions are made.

AI Is A Co-Pilot, Not A Pilot

It processes faster and more consistently than a human — and fails in ways a human would catch instantly. A person reviews the output before it acts, always.

What The Evidence Says

The fear: AI adoption means fewer jobs. A study of 21,559 US firms tracking real, sustained AI spending from Jan 2021 to Feb 2026 says otherwise.

• 10.2% headcount growth at heavy AI adopters, two years after adoption
• $33 vs $3 per employee per month is the real dividing line — intensity beats mere presence
• +12% growth in entry-level headcount among heavy adopters — the opposite of the expected junior-job squeeze

Three Things Worth Remembering

• Intensity beats presence — how much you invest, not whether you use AI
• Entry-level roles grew, not shrank, among serious adopters
• Gains concentrate among organisations already investing seriously

Mindset Shift

From "will AI replace me" to "what can I do now that I couldn't before."

The Equalizer

“AI is the greatest technology equalizer we've ever seen. It lifts the people who don't understand technology.”Jensen Huang, CEO · NVIDIA · CNN Fareed Zakaria GPS, July 2025

What Comes Next

Keep using it on real tasks this week. Every prompt you refine on real data makes the next one better — the skill compounds.

You Do Not Need To Code

You need to be precise. Precision is the skill that transfers — to any tool, any AI model, any problem.