Part One — The Vision

Chapter 1

The AI PowerShift
in Education

Why the most important thing schools can teach right now is not a subject — it is a capability. And why market intelligence is the perfect laboratory for building it.

The Shift Nobody Planned For

In the space of roughly thirty-six months, artificial intelligence went from a specialist topic in computer science departments to the most consequential general-purpose technology in human history. That transition did not come with a curriculum. It did not come with teacher training. It did not come with a textbook that told educators what to do.

What it came with was a question — the same question being asked in boardrooms, government ministries, and school staffrooms simultaneously: what do we actually teach now?

The comfortable answer has been to add AI literacy modules. Teach students what a large language model is. Show them how to write a prompt. Explain the difference between narrow AI and artificial general intelligence. Run a workshop on AI ethics. Tick the box. Move on.

This book argues that the comfortable answer is wrong — not because it is inaccurate, but because it is insufficient. Understanding AI is not the same as being able to use AI to build something real. And building something real is the only thing that will matter to the students sitting in your classrooms today when they enter a workforce that has been restructured around AI capability.

The question is not whether students understand AI. The question is whether students can direct AI — as a tool, as a collaborator, as an infrastructure — toward the production of something that has value in the world.

— The Central Thesis

The shift that nobody planned for is not the arrival of ChatGPT or Gemini or Claude. Those are symptoms. The underlying shift is that the boundary between knowing and doing has collapsed. For most of human history, you needed to know a great deal before you could do anything sophisticated. A student needed years of medical training before they could diagnose a patient. A programmer needed months of practice before they could ship working software. An analyst needed years of market experience before their predictions had any credibility.

AI has compressed that gap to a degree that the education system has not yet absorbed. A motivated student with no coding background can now direct an AI system to build a functional market intelligence pipeline in five weeks. We know this because it happened. In a university classroom. In Singapore. In 2026. This book is, in part, the documentation of that event.

Three Phases of AI in Education

To understand where we need to go, it helps to name where we have been. AI's arrival in education has moved through three distinct phases, and most institutions are still somewhere in the middle of Phase Two.

Phase One — Panic and Policy (2023–2024)

The first response to ChatGPT in educational institutions was largely defensive. The concern was cheating. Students would use AI to write their essays. Assessments would become meaningless. Academic integrity would collapse. Policies were written. Detection tools were deployed. Some institutions banned AI outright. Others added AI statements to their assessment briefs.

Phase One was understandable. It was also, in retrospect, almost entirely the wrong response — not because academic integrity does not matter, but because the energy was spent fighting the technology rather than asking what the technology made possible.

Phase Two — Integration and Literacy (2024–2026)

The second phase has been characterised by integration. AI tools are now permitted in many classrooms. Prompt engineering has become a taught skill. AI literacy modules have appeared in curricula. Educators have started using AI to generate lesson plans, give feedback, and produce resources.

This is progress. But Phase Two is still fundamentally about teaching students to use AI as a consumer tool — to ask it questions, to get it to write things, to use it as an enhanced search engine. The student remains a passive beneficiary of AI capability. They are not yet building with it.

Phase Three — Construction and Deployment (2026 onwards)

Phase Three is where this book lives. It is the phase in which students are not just using AI but directing AI systems to build things that have ongoing, automated, real-world utility. The distinction is not subtle. A student who asks ChatGPT to summarise a news article is using AI. A student who builds a pipeline that automatically collects market data every Friday, sends it to three competing AI models, synthesises their outputs, applies a human intelligence override, seals a prediction, and scores its own accuracy the following week — that student is deploying AI. They are an architect, not a user.

Phase Three in Practice

In TR2 2026, eleven students with no software engineering background reached Phase Three in five weeks. They built an automated system. It ran itself. It scored its own predictions. It opened its own Pull Requests. The educator's job was not to teach them to code — it was to design the exercise that made coding necessary, and to provide the scaffolding that made it achievable.

The Capability Gap

There is a widening gap between what the most capable AI-enabled students can produce and what the curriculum expects of them. This gap is not about intelligence or effort. It is about whether students have ever been given a problem that required them to build a system — not answer a question, not write an essay, not pass a test, but actually build something that works in the world beyond the classroom.

The students who will matter most in the next decade are not the ones who can describe how a large language model works. They are the ones who can look at a problem — in business, in finance, in healthcare, in logistics — and ask: what system could I build, using the AI tools available to me, that would solve this automatically?

That is a learnable skill. It requires exposure to the right problems at the right level of difficulty. It requires scaffolding — structure that makes the problem tractable without making it trivial. And it requires educators who have built something themselves, because you cannot teach construction if you have only ever observed it.

That last requirement is what Part Two of this book addresses directly. Before you teach students to build a market intelligence pipeline, you build one yourself. Not a copy of someone else's work — your own version, from scratch, one step at a time. By the time you finish Part Two, you will have a working system running under your own GitHub account. That experience changes how you teach. Permanently.

Why the Stock Market

The choice of the stock market as the domain for this exercise is not arbitrary. It is deliberate, and the reasons compound on each other in useful ways.

Real Data, Available Immediately

Market data is free, current, and unambiguous. Every Friday evening, the S&P 500, the Nasdaq 100, and the Russell 2000 close at a specific price. That price is a fact. A student's prediction was either directionally correct or it was not. There is no subjectivity, no partial credit, no interpretation. In an educational context where we want students to build systems that produce verifiable outputs, this is enormously valuable. The market is the most honest grader in the room.

Multiple Agents, Multiple Perspectives

A well-designed market prediction exercise requires students to think from multiple angles simultaneously — seasonal patterns, macroeconomic context, technical price signals, and the synthesis of multiple AI model outputs. Each of these is a distinct analytical mode. Each maps to a distinct Scrum role. Each produces a distinct output that can be validated for format and quality. The multi-agent structure of the market intelligence exercise is not a complexity imposed on students — it is a complexity that reflects how real analytical systems actually work.

Commercial Relevance

What students build in this exercise is not a toy. The architecture — data collection, multi-agent analysis, LLM synthesis, human override, automated scoring, CI/CD pipeline — is directly analogous to the systems used by quantitative hedge funds, algorithmic trading desks, and market intelligence services. A student who completes this curriculum has built, at prototype scale, something that exists in commercial form. That is not a metaphor. It is the basis for Chapter 18 of this book.

Intrinsic Motivation

Students care about markets. Not all of them, and not equally — but the combination of real money, real uncertainty, and the possibility of being right when three AI models are wrong produces a level of engagement that purely academic exercises rarely achieve. The Human Score Wild Card — the component of the system where a student applies their own analytical judgment to override or confirm the AI consensus — generates genuine intellectual investment. Students argue about it. They research it. They come to class with views. That is the point.

Real Student Evidence — TR2 2026

What Happens When the Human Score Matters

In Sprint 4, the three mandatory LLMs — ChatGPT, Gemini, and DeepSeek — all predicted a positive week for the S&P 500. The Human Score Analyst on Team 1 disagreed. They had identified a macro signal — a specific Fed communication pattern — that they believed the LLMs had underweighted. They applied the Wild Card override, documenting their reasoning in data/human/human_score_W24.md. The result: their prediction was more accurate than the LLM consensus. The delta engine scored it automatically the following week. The team's calibration log now contains a documented case study of human judgment outperforming AI consensus. That is not a classroom exercise. That is professional analytical practice.

What Students Actually Built

Before making any claims about what this curriculum can produce, it is worth being precise about what the students in TR2 2026 actually built. Not what they learned in the abstract — what they built, concretely, in five weeks.

Team 1 — eleven students, none of them computer science majors — produced the following:

This was not the work of one gifted student. It was the work of eleven students, each occupying a specific Scrum role, each contributing through a defined workflow of branches, Pull Requests, and CI validation. The system was built collectively, incrementally, and in public — every decision visible in the commit history.

The other four teams built at varying levels of sophistication. Two had working automated pipelines by Sprint 5. One built a TypeScript dashboard named TradeKyaMal. One went from zero automation to a working GitHub Actions workflow in seven minutes during Sprint 5. All five teams sealed prediction files every week and tracked their accuracy. All five teams had every member visibly contributing on GitHub by Sprint 4.

This is what the right curriculum produces. Not essays about AI. Systems that use AI.

The Central Claim of This Book

This book makes one central claim, and everything that follows either supports it or shows you how to act on it:

Any educator — regardless of their technical background — can design and deliver a curriculum that produces AI-powered, commercially relevant student projects within a single semester. The prerequisite is not coding skill. It is the willingness to build something yourself first.

— Professor Dr. Tan

The qualifier — build something yourself first — is not a throwaway condition. It is the central pedagogical requirement. You cannot teach students to navigate unfamiliar technical territory if you have never navigated it yourself. You cannot assess their work with confidence if you do not understand what the work involves. You cannot answer their questions, anticipate their failures, or recognise their breakthroughs if you have not experienced the same process.

Part Two of this book is where you build something yourself. It starts at absolute zero — creating a GitHub account, understanding what a terminal is, writing your first file. It ends with a working automated pipeline running under your own account, validated by CI, producing outputs every week without your intervention.

That journey — from zero to working system — is the same journey your students will take. Having made it yourself, you will teach it differently. Better. With the specific authority that comes from personal experience rather than theoretical knowledge.

The chapters that follow are designed to make that journey as direct and frustration-free as possible. Every step is documented. Every command is explained. Every expected outcome is described so you know whether what you are seeing is correct. And every concept is connected to what the TR2 2026 students actually did, so the pedagogy is grounded in demonstrated results rather than theoretical possibility.

The AI PowerShift in education is not a future event. It is happening now, in classrooms around the world, with students who are building things their teachers have never imagined. The only question is whether those teachers are positioned to guide that work — or whether they are watching it happen from the outside.

This book is for the educators who choose to be on the inside.

What to Do Before Chapter 2

Nothing technical is required yet. But between now and Chapter 2, spend twenty minutes observing something: look at how a student in your current class uses AI. Are they consuming its outputs, or directing it toward a specific goal they defined? The difference between those two modes is the difference between Phase Two and Phase Three. By the end of this book, you will know how to move every student in your class from the first mode to the second.

← Back to Home