AI Makers: AI Camp Syllabus for Ages 15+

AI Makers: Ages 15+

The pitch for campers: You've used AI to get answers. Now build AI that gets things done: an agent that runs part of your life, a real project that shows what you're capable of, and a personal rulebook for using AI without losing your edge. No coding required. If you want code, see AI Architects.

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

• Hands-on with consent. Campers 15–17 may use ChatGPT with parental consent, ^[1] and parents can link accounts for teen protections. ^[2] Claude.ai stays off-limits until 18. ^[3] The camp provisions supervised accounts or collects consent before Day 1.

• The habits are already forming. Pew found that teens aged 15–17 are more likely than 13- to 15-year-olds to use ChatGPT, and about three in ten teens use chatbots daily. ^[4] The goal is to turn passive use into deliberate building, and to protect the learning that makes them good at things.

• Coach, not crutch, made explicit. Campers read the summary of two studies and argue about why they disagree. One found that unrestricted GPT-4 hurt high-school math learning once it was removed. ^[5] The other found that a well-designed AI tutor doubled learning gains for Harvard physics students. ^[6] The answer, design, becomes the camp's rule for every agent they build.

• UNESCO level: Apply into Create. ^[7]

Ratio: at most 1 adult to 12 campers. Devices: laptop with a modern browser. Platform choice: any general chatbot that supports custom instructions or custom assistants, plus a no-code automation tool, chosen by Colearn for its data terms.

Course 1: Main Character Energy

Maps to the original AI Mastery for Living Life: Personal, Social, and Work/School agents.

Big question: If you could hand an agent one part of your life, which part, and what would you never hand over?

Campers will be able to:

• Explain the difference between a workflow (steps fixed in advance) and an agent (the model chooses its next step), and when the simpler one is better. ^[8]

• Build three working custom assistants or automations, one per life area, each with written instructions, a knowledge file, and a Check-in.

• Evaluate each one with a test set of 10 realistic requests, including at least 3 meant to break it.

• Write a one-page "AI operating manual" for themselves.

Sources for Table: ^{[9] [10]}

Capstone: three working agents, a test log for each, and a personal AI operating manual.

Course 2: Spike Season

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

Big question: What could you ship in five days that a college, an employer or a community would actually care about?

Campers will be able to:

• Pick a spike and scope a five-day project with a real audience.

• Use AI as research assistant, prototyper and critic while keeping authorship. The US Copyright Office's January 2025 report holds that prompts alone don't make you the author; human creative choices do. ^[11]

• Cite AI-assisted research properly, and verify every claim AI gives them against a primary source.

• Publish the project somewhere real.

Sources for Table: ^[12]

Capstone: a published project, its contribution log, and a cited one-page brief.

Course 3: Signal Over Noise

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

Big question: How do you use AI for your inner life and your relationships, and where is the line?

Why it matters: Common Sense Media recommends that no one under 18 use AI companions. ^[13] Its testing found that Claude, while it has safety strengths, is not safe or reliable for teens' mental health or emotional support. ^[14] Meanwhile, the Surgeon General has named loneliness a public-health crisis. ^[15] Campers investigate the problem as researchers, then build what should exist.

Campers will be able to:

• Analyse how engagement-driven design works in AI products and social feeds.

• Build a "signal" agent that filters information toward what they care about and away from doom-scrolling.

• Write and defend a personal policy on AI for emotional topics: what it's okay for, and who to go to instead.

Sources for Table: ^{[16] [17] [3] [18]}

Capstone: two working agents plus a written, defended personal AI policy.

Course 4: Memory Palace

Set up your harness, build your second brain. New for ages 12 and up, and it sits underneath all three original courses: it's what turns a generic chatbot into an agent that knows your projects, your goals and your rules, without knowing anything it shouldn't.

Big question: How do you give an agent exactly the right context, no more and no less, while keeping the knowledge yours?

Why it matters (the version for parents): The people who get the most out of AI right now aren't the best prompt-writers. They're the ones who set up the system around the model. Andrej Karpathy, a founding member of OpenAI, calls the core skill "the delicate art and science of filling the context window with just the right information for the next step." ^[19] It matters because a model's attention is finite: Anthropic describes an "attention budget" that every added token draws down, ^[20] and Chroma's tests of 18 models found reliability slipping as inputs grow longer. ^[21] Anthropic's own engineers compare an agent to a project "staffed by engineers working in shifts, where each new engineer arrives with no memory of what happened on the previous shift." ^[22] The harness (instructions, memory files, skills, tools and compaction) is how one shift hands over to the next.

The second brain is the other half. It's a personal library of notes, written and owned by you, organised with Tiago Forte's CODE method (Capture, Organize, Distill, Express) and PARA folders. ^[23] It's kept as plain files that outlive any app. ^[24] A well-kept second brain makes you sharper, and it's also the best context an agent can be given.

Tools at this age:

• Hermes Agent (open source, MIT-licensed, works with any model provider). ^[25] Each camper gets their own Colearn-provisioned sandbox, uses the command line only, keeps command approval on and messaging connections off, and runs on Colearn's model keys with spending caps.

• Second brain: Logseq (open source, local Markdown) ^[26] or any Markdown editor. Obsidian's terms limit use to adults, ^[27] so it's for campers who are 18, or instructor demos. Every vault is plain Markdown and moves freely between tools.

Campers will be able to:

• Set up and tune an agent harness: persona, workspace instructions, memory, and one custom skill.

• Build a second brain with CODE and PARA, including a distilled synthesis they wrote themselves.

• Audit their own context: count what the harness loads every session, and cut what doesn't earn its place.

• Carry work across sessions with compaction (/compress) and structured notes, and test what survives.

• Write a skill using progressive disclosure: the agent always sees a short name and description, and loads the full instructions only when they're needed. ^[28]

Sources for Table: ^{[29] [21] [30]}

Capstone: an exported, tuned Hermes setup; a vault of 30+ notes with one distilled synthesis; a context-rot lab chart; and one working skill.

Assessment for this band

Apply the shared Four Lenses rubric from Program Design. Evidence expected at this age:

• Build: the agent works for someone other than its maker.

• Judgment: a 10-request test log with failures fixed or documented.

• Purpose: a project brief that names a real audience.

• Humanity: a contribution log and a personal policy that name real people.

• Judgment (Course 4): a context budget the camper can defend line by line, and memory files with nothing in them that shouldn't be there.

Instructor prep

• Confirm consent forms and account provisioning a week before camp. ^[1]

• Pre-build one example agent per course, including deliberate flaws for campers to find.

• Prepare the two-study reading pack (Bastani et al., Kestin et al.) as one-page summaries. ^{[5] [6]}

• For Course 4: provision one Hermes Agent sandbox per camper, with command approval on, messaging connections off, and a capped model key. ^[25] Wipe or export each camper's memory, skills and vault at the end of camp, and tell families which.

Sources