The Final Sprint

Break It Before They Do, and the Pitch

Overview

# Overview

Curriculum.Units

Testing a product adversarially, remediating and pitching it. Students hunt defects, red-team their own agent, keep a public defect log, and deliver a costed live pitch to a room that did not build the product.

# Objectives

Curriculum.Units
  • 01 By the end, you can Test your own product to find defects before users do — click every button, read the console, watch a stranger.
  • 02 By the end, you can Attack your own agent — flood it, push it off-topic, try to expose its prompt — and log everything that gets through.
  • 03 By the end, you can Keep a public defect log, claiming no fix without a logged entry.
  • 04 By the end, you can Combine both into a live pitch that runs the product for a room that did not build it and answers the money questions.

# Sessions

Curriculum.Sessions
  1. Testing Your Own Product
    Applied craft
    • Simulation Break It Hour Four testers each try to break the product from a different angle and write down every problem they find.
    • Simulation Fresh Eyes Test Give the product to someone new, hand them one task, stay silent, and note every spot where they get stuck.
  2. Attacking Your Agent
    AI concepts
    • Simulation Emoji Flood Stress Test Each team fires a thousand messages and five thousand emojis at their own AI assistant and watches whether it slows down, breaks, or stays polite.
    • Game Push It Off Topic Teams take turns trying to trick their assistant into talking about something it should refuse, scoring a point each time it goes off the rails.
    • Worksheet What Got Through Log Each team writes down every message that broke the rules or slipped past the assistant so they can see the pattern of weak spots.
  3. Generating Pitch Assets With Codex
    Tool teaching
    • Build Build The Pitch Deck Use an AI tool to turn the project into slides, then rehearse a five minute pitch where every role speaks.
  4. The Pitch and Live Demo
    Deliverable prep
    • Role-play Four-Pillar Pitch Rehearsal Each team member practices speaking their part of the five-minute pitch covering the four topics while a stopwatch keeps them on time.
    • Role-play Money Questions Hot Seat One team sits in front of the class and answers tough questions about their product on the spot while the audience plays investors.
    • Demo Live Demo Dry Run Each team runs their product live in front of another team to catch anything that breaks before the real pitch.

Applied · App Development

# Knowledge & Skills

ACA-Applied
K Knowledge · what you know
A defect caught while the product is being built costs little to fix, and the same defect found after release costs more. The cost of fixing a defect rises at each stage: design, build, test, and release.
Usability is how a product feels to use, not how it looks, and a good product explains itself so the user does not have to work it out. Three failures to look for are: they do not know what to do first, they cannot tell whether it worked, and they cannot recover from a mistake.
Five usability checks apply: the system's status is visible, it matches how the real world works, it prevents errors, users can recover from mistakes, and it stays consistent.
The DIY test means handing the product to someone who has never seen it, giving one task, and saying nothing, where every hesitation counts as a defect.
An API (the door that lets programs talk to your product) has three risks: a stolen key, too many requests at once (flooding), and bad input.
Anyone can read the front end (the code sent to a user's browser), so keys stay on the server and a proxy (a go-between program) exists so the browser never sees the key. You test this by searching the page source for the key.
Every defect is logged in the open and marked when fixed, and a fix that was not logged leaves no proof it happened.
Never accept a fix you did not read: write the problem as one sentence, let the tool propose the change, read the change, and test again.
The pitch is given by all four roles, and the product is shown working live.
S Skills · what you can do
Present a five-minute pitch on four topics: the vision, the technologies used to build the product, the brand and customer, and the business case, with every role speaking. point w3d5-s2-p1; evidence w3d5-e5
Answer the money questions: pricing, costs, competitors, and what you would do with a $500k investment. point w3d5-s2-p2; evidence w3d5-e3
Show the product working live in front of a room that did not build it. point w3d5-s2-p3; evidence w3d5-e6
T Techniques · named subskills
All-Roles Delivery Every team role speaks during the pitch, not just one presenter.
Answer-for-the-Money Drill Be ready on pricing, costs, competitors, and what you'd do with a $500k investment.
Four-Pillar Structure Vision, tech stack, brand & customer, business case — the fixed shape of a final presentation.
Role-Voiced Reporting Each team role narrates their own part of the build, not a single spokesperson.

AI

# AI Concepts

ACA-AI.Concepts
Aligned with AI4K12 · Five Big Ideas in AI View framework →
AI4K12 · Five Big Ideas in AI

Pitch Built From Real Product

Codex can generate the pitch deck, an infographic, and a promo pack from the product's own summary file (the README), its code, and its numbers, so the presentation is built from the real product and not from a blank slide.

AI4K12 · Five Big Ideas in AI

Cut Slides to Four Topics

AI-generated slides include extra text that is not needed (filler), and the deck becomes clear only once it is cut to the four topics and the live demo.

AI4K12 · Five Big Ideas in AI

Four Named QA Roles

There are four named QA roles: prompt and code engineer, UX specialist, security and API tester, and QA lead and scribe. QA (checking a product's quality) is a real job on every team.

AI4K12 · Five Big Ideas in AI

Prompt Injection Overrides System

Prompt injection is when a user's input overrides the system prompt (the agent's own hidden instructions), and it is the top-ranked risk for AI language model applications.

AI4K12 · Five Big Ideas in AI

Six Red-Team Probes

The red-team method has six probes: flood it, push it off topic, tell it to ignore its instructions, ask it to reveal its system prompt, feed it junk, and ask it to do something it should refuse.

# AI Competencies

ACA-AI.Competencies
Aligned with UNESCO · AI Competency Framework for Students View framework →
Human-Centred Mindset
4.1.1

Human Agency

Understanding that AI systems are tools created by and for humans, and that humans must maintain control and decision-making authority.

AI Techniques and Applications
4.3.4

Iteration and Feedback

Using systematic feedback and improvement processes to refine AI systems and collaborate in their ongoing development.

# AI Skills

ACA-Tools.Skills
Aligned with Derived from course activity
Derived from course activity

Prompt Engineering for AI Agents

Iterating a prompt/system instruction until an AI agent's responses stop sounding generic and actually reflect the specific product or context. Evidence: "Does this information sound generic? Or does it sound actually tailored to the website?" (app-development, w3d3).

Derived from course activity

Deployment & Secret Management

Setting up a dev environment and deploying live while keeping API keys out of source, held as environment variables. Evidence: "SETTING UP A DEVELOPMENT ENVIRONMENT: 1. Make a folder on your laptop..." and "holding the API key as an environment variable" (api/app-development).

# Competency Goals

ACA-AI.Goals
Aligned with UNESCO · AI Competency Framework for Students View framework →
UNESCO · AI Competency Framework for Students
CG4.3.4.2

Technical Skills with Feedback

Using performance data, user feedback, and testing results to guide technical improvements and refinements.

UNESCO · AI Competency Framework for Students
CG4.2.1.1

Accountability Through Lifecycle

Ensuring human oversight and responsibility at every stage of an AI system's development, deployment, and use.

UNESCO · AI Competency Framework for Students
CG4.3.2.1

Ethics Through AI Lifecycle

Embedding ethical review, consideration, and decision-making at each stage of AI development from conception to deployment.

UNESCO · AI Competency Framework for Students
CG4.2.1.3

Competencies to Steer AI

Building capacity for people to audit, understand, challenge, and control AI systems they depend on.

UNESCO · AI Competency Framework for Students
CG4.2.4.1

Evaluation Frameworks

Using systematic methods to assess AI system performance, fairness, safety, and alignment with intended purposes.

UNESCO · AI Competency Framework for Students
CG4.1.3.4

Human-Centred Design Considerations

Learning to design AI systems with human needs, values, limitations, and diverse perspectives in mind.

UNESCO · AI Competency Framework for Students
CG4.3.2.2

Critical Attitude to AI Ethics

Maintaining skepticism and questioning about ethical claims, practices, and frameworks in AI development.

UNESCO · AI Competency Framework for Students
CG4.2.2.2

Recognising AI Hazards

Identifying potential harms, risks, and failure modes that AI systems can cause to individuals and communities.

UNESCO · AI Competency Framework for Students
CG4.1.1.2

Necessity of Human Control

Understanding that critical decisions should not be entirely delegated to AI systems without human oversight.

UNESCO · AI Competency Framework for Students

Test Before a User Does

Test your own product before a user does: click every button, read the console (the browser's technical error report), and watch a stranger use it.

UNESCO · AI Competency Framework for Students

Attack Your Own Agent

Attack your own agent (the AI assistant you built) with a thousand messages and five thousand emojis, push it off topic, and log what gets through.

UNESCO · AI Competency Framework for Students

Log Defects in the Open

Log every defect in the open and mark it when it is fixed.

UNESCO · AI Competency Framework for Students

Cut Slides, Remove Filler

Cut AI-generated slides down to the four topics and remove the filler.

UNESCO · AI Competency Framework for Students

Generate Deck From Repo

Generate a pitch deck (a set of presentation slides), an infographic, and a promo pack from the repo (the folder that holds the project's code) using Codex (an AI tool that writes and edits code).

Deliverable

# The Final Sprint

Projects.Briefs

Find defects in your own product and agent before users do, then pitch the product and show it working live. A public defect log, plus a five-minute pitch on four pillars, with the product shown working live and a Codex-generated deck, infographic and promo pack, all cut to the four pillars.

🏆 Capstone deliverable

Transfers to: Trying to break your own work before someone else does it for you.

See past cohorts ship this in the Showcase →

# Assessed toward these aims

Projects.Rubrics

This deliverable is where students demonstrate

  1. E Test the product adversarially, remediate, and pitch it
    Look out for
    • Defect hunting: Every button is clicked. The console is read. A stranger is watched doing one task. Every defect found is written as one clear sentence. This ties to: test your own product.
    • Agent attack: The agent is flooded with messages, pushed off-topic, told to ignore its instructions, and asked to reveal its system prompt. Everything that gets through is logged. This ties to: attack your own agent.
    • Public defect log: Every defect is logged in the open and ticked off when fixed. No fix is claimed unless it has a logged entry. This ties to: log every defect in public.
    • Live demo and pitch: The product runs live in front of a room that did not build it. All four roles speak. The pitch covers the four pillars in five minutes. This ties to: present the pitch and show the product working live.
    • The money answer: Pricing, costs, competitors and a plan for a $500k investment are all answered with real numbers. This ties to: answer for the money.