AI Builders: AI Camp Syllabus for Ages 12–14

AI Builders: Ages 12–14

The pitch for campers: Everyone your age is using AI. Almost nobody is building with it. This week you'll design a crew of agents for your actual life, ship an app or project in the area you're most into, and learn the skill that matters most: telling when AI is right, wrong or just confidently making stuff up.

Part of AI Camp. Shared principles, the Agent Card, the Four Lenses rubric and the tools-by-age table live there.

Why this age band is built differently

• The age line runs through the middle of the room. A 12-year-old is below ChatGPT's minimum age. A 13- or 14-year-old may use it, but only with parental consent. ^[1] Since mixed cohorts follow the youngest camper, the default is that the instructor runs chatbot demos and campers build with tools that don't need a personal chatbot login. 13+ cohorts with signed consent can unlock supervised hands-on chatbot time.

• They're already using it. In Pew's autumn 2025 survey, more than half of US teens aged 13–17 had used chatbots for schoolwork (54%) or to search for information (57%). ^[2] This band is the moment to build judgment before habits set.

• The right ready-made curriculum exists. Experience AI, from the Raspberry Pi Foundation and Google DeepMind, is designed specifically for ages 11–14. ^[3] MIT App Inventor lets campers ship real phone apps with blocks, including AI features. ^[4]

• UNESCO level: Apply. Campers move from knowing what AI does to using it deliberately for a purpose they choose. ^[5]

Ratio: at most 1 adult to 12 campers. Devices: laptop, plus a phone or tablet with the MIT AI2 Companion app for App Inventor builds.

Course 1: Level Up: Build Your Agent Crew

Maps to the original AI Mastery for Living Life.

Big question: What would a team of AI helpers for your life look like, and how do you stay the boss?

Campers will be able to:

• Explain the difference between an AI that answers and an agent that acts: one that has a goal, uses tools, and decides its next step. ^[6]

• Write a complete Agent Card for three agents (Personal, Social, School) and build one of them.

• Run a "bluff test" that catches a chatbot stating something false with confidence.

• Design a Check-in that keeps a human in charge.

Capstone: one working App Inventor agent plus three Agent Cards, with a written Check-in for each.

Course 2: Spike Lab

Maps to the original AI Mastery for the Work/School World (STEM, MAD, ELC).

Big question: What's the one thing you could get unusually good at, and how can AI speed you up without doing it for you?

A spike is one area where you go deep instead of being a little good at everything. Campers sample all three worlds, then commit to one.

Campers will be able to:

• Complete the full AI project cycle on a problem in their spike area: question, data, evaluate data, train, evaluate model, improve. ^[9]

• Explain who owns AI-assisted work. The US Copyright Office says prompts alone aren't enough for copyright; human creative contribution is what counts. ^[10]

• Present a project with a clear "what I did vs. what the AI did" breakdown.

Capstone: a working spike project plus the "what I did vs. what the AI did" slide.

Course 3: Squad Goals

Maps to the original AI Mastery for Personal & Social Life.

Big question: How do you use AI to be a better friend, family member and teammate, without letting it replace any of them?

Why it matters: Common Sense Media rates social AI companions an unacceptable risk and says nobody under 18 should use them. ^[13] In its 2025 national survey, it found that many teens use them anyway. ^[14] This course doesn't lecture. It has campers design the opposite: agents built to send you back toward people.

Campers will be able to:

• Describe how AI companions are designed to keep you engaged, and why that's a business choice, not friendship.

• Build a "people-first" agent whose success is measured in real-world connection.

• Draft a personal AI and screen-time plan with their family.

Capstone: a people-first agent app plus a family AI plan draft.

Course 4: Second Brain, First Agent

New for ages 12 and up. It sits underneath all three original courses: it's the setup that makes the Personal, Social and Work/School agents actually know what they need to know.

Big question: What does an AI actually "know" about you, and who gets to decide?

Why it matters (the version for parents): A language model doesn't remember you. In any conversation, everything it knows sits inside a limited context window, and the software wrapped around the model, the harness, decides what goes in: instruction files, memory, skills and tools. Anthropic's engineers describe context as "a finite resource with diminishing marginal returns," with an "attention budget" that gets spent as more is added. ^[17] Researchers at Chroma tested 18 leading models and found performance gets less reliable as inputs grow, even on simple tasks. ^[18] So the skill isn't giving AI more; it's giving it the right things.

The second half of the idea is older than computers. In 1945, Vannevar Bush imagined the memex, a desk that would hold all of a person's books, records and notes as "an enlarged intimate supplement to his memory." ^[19] Today a second brain is a folder of plain-text notes you write and own: "files you can control, in formats that are easy to retrieve and read." ^[20] That also happens to be exactly what agents read best. Kids who build one now get a thinking tool for life and the raw material for every agent they'll ever set up.

Tools at this age:

• Hermes Agent from Nous Research is an open-source (MIT-licensed) agent that keeps memory across sessions, builds its own skills, and runs on any model provider. ^[21] At this age the instructor runs it on a camp machine with Colearn's model keys and its messaging connections switched off. Campers write the files that shape it and watch what changes.

• Second-brain app: Obsidian's terms of service limit use to people who have reached the age of majority, ^[22] so campers build their vaults in Logseq, an open-source (AGPL), local-first notes app that stores plain Markdown, ^[23] or in any Markdown editor. Instructors demo in Obsidian. Because every vault is plain Markdown, it opens in any of these tools, now or later.

Campers will be able to:

• Explain context window, harness and memory in their own words, and why dumping in more information can make answers worse.

• Build a starter second brain: 20 or more linked notes, written in their own words, organised with PARA (Projects, Areas, Resources, Archives). ^[24]

• Write the three files that shape an agent: a persona (SOUL.md), house rules (AGENTS.md), and a profile of themselves (USER.md), and show how editing them changes the agent's behaviour.

• Write a never-store list: what an agent must never remember about them.

Sources for Table: ^{[25] [26]}

Capstone: a personal vault of 20+ linked notes, three harness files with a never-store list, and a before-and-after comparison showing what changed when the context changed.

Assessment for this band

Apply the shared Four Lenses rubric from Program Design. Strong work at this age shows:

• Judgment: a documented Bluff Test, or a red-team fix.

• Humanity: a Check-in that names a specific person and a specific situation.

• Build: an app that runs on a real phone.

• Purpose (Course 4): a vault of notes written in the camper's own words, and a USER.md that fits the budget because the camper chose what matters.

Instructor prep

• Experience AI lesson materials for ages 11–14 cover the "how models learn" sections. ^[3]

• Set up App Inventor class accounts and test the AI extensions before Day 1. ^[4]

• Collect parental-consent forms before any 13+ hands-on chatbot time. ChatGPT supports linked parent accounts for teens. ^[27]

• For Course 4: install Logseq on camper machines, and set up one instructor-run Hermes Agent on a camp machine with messaging connections off and a model key that has a spending cap. ^[21] Clear Hermes's memory folder between campers.

Sources