FutureMasters · The 12-Week Curriculum · North Star Edition

The scarcest resource of the AI century is a human who knows what they want to make.

AI now produces output for free. What no AI produces is a kid with direction: one who knows what they are passionate about, knows what they want to exist, can command the tools to build it, can catch the machine being wrong, and can stand in front of real humans and make them see it too: pulling people into their vision until their lane is different because they showed up.

• The 11-year-old obsessed with fishing games who ships a playable one.
• The 13-year-old who turns a year of worldbuilding lore into an animated teaser.
• The 14-year-old military history encyclopedia who builds an interactive supply-line map.
• The 16-year-old novelist who publishes chapters strangers actually read.
• The 17-year-old who grows a 90-member builder community and runs it like a founder.

• 12 weeks. Live. High-school-age kids (14+, framed soft: adolescents, and adults benefit too). Customer is the parent; the product is for the kid.
• The flywheel: Spark to Tools to Ship to Share. Every week lives in one phase of the loop, and every kid runs the full loop at least twice.
• Three live layers, every week. One: the expert session (1 hour, one-to-many, recorded). The god-tier mentors: world-class leaders in their space who show exactly how they apply AI in their world. Every session ends with a challenge.
• Two: demo day (small groups, time-zone slotted). Kids show prototypes, links, and work in progress to mentors and peers. Sharing is not an afterthought; it is part of the loop.
• Three: the crew. A peer community of builders where shipping is normal and standing up to share your work is the culture.
• Every age group teaches down to the one below as they learn. Learn it this week, teach it next week: the teach-down is part of the loop, not an extra.
• JARVIS is the constant. A persistent AI companion that knows the kid's spark, projects, and history. JARVIS is specifically tuned to guide kids through this curriculum and this way of thinking: amplifying their spark and driving the flywheel with them, week after week.
• Every project ends with the kid standing up and sharing it. The Closed-Laptop moment is the program's atomic unit of proof: explain what you made, how you verified it, and what magic was only yours, laptop closed. But the point is not defending work from attackers; it is propagating a vision. Owning the work so completely that they can hand it to a room and watch it land. If they can't explain it, they didn't make it. When they can, people follow it.

Sixteen North Stars govern everything FutureMasters teaches. The twelve weeks are those stars turned into reps.

• Live expert session (1 hr): a mentor, a world-class leader in their space, talks about their own spark and shows exactly how they apply AI in their world; issues the week's challenge.
• Build blocks (2x 60-90 min): kid + JARVIS working the week's challenge; every week they get something real out the door, and the streak is celebrated.
• Demo day (small group): show the work to mentors and peers: link, prototype, or progress; take questions.
• Closed-laptop rep: one laptop-closed explanation per week, minimum.
• Teach-down: each age group passes what it just learned to the group below.
• Debrief into JARVIS: what was delegated, what was kept, where the model was wrong, where the magic only I could add went in.

Not a generic problem off a list. The one you already care about. Nail the why in these two weeks, and every week after pays compound interest on it.

• Onboarding conversation with a coach, one-on-one, safe environment: what does this kid actually care about? Discover, amplify, codify the spark. The Spark Journal and Constellation map in the app hold it from day one, and JARVIS learns every entry: the curriculum starts adapting to this kid from the first session.
• Agency, named on day one: what it means to act on the world instead of waiting for an assignment, and why it matters more every year the tools get stronger.
• The deeper bet: your spark is not decoration. It is strategic. The world is filling with tools that can answer, draft, code, draw, search, and simulate. That makes caring about a thing more valuable, not less. The kid who cares will ask better questions, hold a higher standard, notice what the generic answer missed, and keep going after the novelty wears off. Universities still mostly train for a world where knowledge was scarce. FutureMasters trains for a world where intelligence is everywhere, and the rare thing is a human with taste, stamina, and a reason to build.
• JARVIS is tuned to the spark from day one; it opens every session asking what they're building toward. Week 1 is also the ingestion week for the dynamic curriculum: the onboarding conversation, the Spark Statement, and the baseline recording all feed JARVIS's picture of who this kid is and what pulls them. From here on, the program is not being delivered to them; it is being shaped around them.
• Baseline Closed-Laptop Defense: kid explains their most recent school or AI project, laptop closed, on camera. This recording is the "before."
• Challenge: The Spark Statement. One page: my lane, why it pulls me, three things I want to exist that don't.
• Demo day 1: introduce yourself with your spark, not your grades.

• The ladder, taught explicitly: Stage 1 is talking TO AI; Stage 2 is working WITH an AI that has hands: search, code, files, self-checks. Once the AI has hands, the kid's job changes from typist to director.
• The director's loop: brief, review, reject, own. Passengers accept first drafts. Directors send them back.
• The Prompt Review dojo (in the FutureMasters app): every mission brief runs through the analyzer; kids see their brief graded on the spot and rewrite until it commands instead of asks. The difference is structural, not stylistic. "Can you tell me about healthy dinners?" gets a generic lecture. "Plan five dinners for a family of four, under 30 minutes each, no seafood, and give me the one grocery list" gets a result. Asking invites the machine to improvise from its brain soup, which is where hallucinations live; commanding pins it to your context, your constraints, your definition of done. And commanding is also where agency starts: the kid who asks stays stuck in the chat window, while the kid who commands has the AI drafting the outreach emails, booking the room, and organizing the event out in the real world.
• Intro to Context: a first taste of the superpower, why the same request lands differently depending on what the machine knows. Full training next week.
• Challenge: The First Mission Brief. Delegate a real task from your lane to JARVIS with a written brief; grade the output against your standard; reject and re-brief until it passes.
• Demo day: show the brief, the rejected draft, and the accepted one, side by side.

• What context is, why the same request lands completely differently with it, how to gather, structure, and preserve it across projects, conversations, and long-running work. The app's Context pages (Gather it, Structure it) are the week's workbench.
• Context management is one of the big unlocks of high-agency AI use, and nobody teaches it. Directors don't just write instructions; they pack the briefcase.
• Structured review is context management in the wild: take messy spoken context, distill it into written edits, organize the pieces, clear the noise, and run the next clean pass. The roadmapping system does the same thing for building: turn fog into an ordered plan, then execute from the plan instead of from memory soup.
• Challenge: The Briefcase. Pack a reusable context briefcase for your lane (goals, constraints, taste references, prior work); run the same task with and without it; measure the difference.
• Demo day: show the with/without diff. Numbers, not vibes.

• The machine lied to you this week. Fluently, confidently, several times. Most people never notice; this week the kid becomes someone who does.
• Live demonstration to open: the coach makes JARVIS produce a beautiful, wrong answer in the kid's own lane, on camera. The room watches confidence and correctness come apart in real time.
• The verification reflex, drilled as a game: source, seam, sanity-check. Speed rounds in the Detective Log: two answers, one poisoned, find it before the timer.
• Trust architecture: what you always verify (numbers, citations, anything you will sign), what you sample, what you never delegate (the final call).
• Challenge: The Lie Hunt. Run a real research task in your lane through JARVIS, then catch and document every hallucination against primary sources. Best catch wins the week. The wrong-log starts here and runs the rest of the program; wrong-log depth is a tracked stat, like a batting average.
• Demo day: present your best catch. "My kid catches AI being wrong" is the brag we're building. The kid who catches the machine deceiving them once never reads AI output the same way again.

• From single prompts to chains, tools, and agent workflows with verification steps built in. The Prompt Review dojo grades every One-Shot Mission.
• Challenge: The One-Shot Mission. Write one complete mission brief for a real multi-step task; grade yourself on one-shot success rate (shipped with zero follow-up corrections).
• Demo day: read your brief aloud; the group predicts where it will fail before you reveal the run.

• The Problem Selection Studio: the crown skill of the Trio That Cannot Be Automated. Scan your domain, map its real problems, interview actual humans, size the opportunities, and pick ONE for your capstone.
• Lane: Build. A working tool, game, or physical thing (code, 3D printing, hardware) real people use.
• Lane: Creative. A substantial work where AI multiplies output but the taste is unmistakably theirs.
• Lane: Research. An original investigation with provenance discipline.
• Lane: Venture. A micro-business with real customers or real no's.
• Lane: Community. A living group, event, or movement the kid founds and runs, measured by participation.
• Lane: Advocacy. A campaign that changes a real decision: a school policy, a local cause, a public argument made with evidence.
• Challenge: The Pitch. Bring mentors and peers into your vision: the lane, the mission, why it matters, and what you will make real. The choice is graded harder than the solution will be.
• Demo day = pitch day. Parents invited. This is the midpoint gate: no capstone starts until the mission has landed with the room.

• Deploy is not a programmer word. Posting the episode, publishing the piece, listing the product, sending the pitch: all deployments. The kid-sized ladder: show one person, show a room, show your community, show the world.
• Week 7 ends with something live: a URL, a published piece, a listed product, a sent pitch. Feedback only exists after deployment.
• Challenge: Ship Zero. Smallest real version of the capstone, public by demo day. Ship cadence tracking starts now and runs to week 12.
• Demo day: proof it met the world. A link is the classic form, and most weeks it is a link. But a 3D-printed part on the table counts. Five humans who showed up to a meeting JARVIS planned counts. A real person changed by the work counts. If it did not touch reality, it does not demo.

• Execution is cheap now; taste is not. Original taste is trainable: reps of judging, comparing, choosing, and explaining why one version is alive and the other is sludge.
• Critique cycles: study the human greats of your lane, the people whose work made you pick this lane in the first place, then edit AI output until it meets the standard they set in you. Which-is-better drills: two nearly identical outputs, make the call and explain why.
• The live question stays open all year: where can AI beat the greats, and where does the human still tower? Beethoven in music. McQueen in fashion. Miyazaki in animation. The point is not worship; it is calibration. The kid studies the greats in the lane their spark chose, then learns what standard their own work is trying to approach.
• Challenge: The 10x Pass. Take the basic creation from Ship Zero and make it 10x better, 10x more unique, and unmistakably yours. Document every taste decision, cut, rewrite, test, and refusal that the machine could not have made.
• Demo day: before/after, with the decision log.

• Agentic AI is the trendy phrase adults throw around. FutureMasters makes it visible enough for a 15-year-old to reason about. An agent is not just a model. It is a mix: model + memory + tools + orchestration. The model thinks, the memory remembers, the tools act, and the orchestration decides what happens next. An agent's personality comes from all four together: change any one of them and the same "assistant" behaves differently.
• Your kid is not trying to become an AI researcher in Week 9. They are learning enough to see the system operate, understand each component's job, and know where to look when the agent surprises them.
• Hype immunity: when a new "best model" drops, run your own evals on your own tasks before adopting the take. The Model Scorecard in the app keeps every kid's evals; JARVIS itself is the standing exhibit: the kid has already watched its engine rotate.
• Challenge: The Model Race. Build the same thing twice with two different frontier models inside the same agent mix. Compare output quality, speed, factuality, taste, failure modes, tool-use behavior, and how much human correction each model needed. The goal is not to crown a universal winner; it is to learn which model wins for your task.
• Demo day: show the mix diagram and the race results.

• The pressure test: here is a problem you have never seen, here is an AI, you have 20 minutes. Every skill from weeks 2-9 in one live room: briefing quality, verification under pressure, recovery from bad output, honest ownership of the result.
• This is the compounding made visible: rooms like this are where directed-AI skill gets judged for the rest of their lives, and rehearsal beats theory every time.
• Example cold brief: Bridge Tender. Concept, load calcs, physical model, destructive test.
• Example cold brief: Disaster Logistics. Plan hurricane supply routes, then handle a closed bridge and fuel shortage.
• Example cold brief: Flow Plateau. A browser extension has 96 weekly active users and flatlined. Find three growth experiments without breaking user trust.
• Example cold brief: Darkborn Teaser Cut. An indie animation world has too much lore. Pick the 60 seconds that make strangers care.
• Example cold brief: Brain Break Policy. A school wants a 10-minute reset between AP classes. Design the survey and the argument that could move a principal.
• Example cold brief: Tobruk Supply Map. A WWII logistics explainer is drowning in facts. Turn it into a clear interactive map for non-experts.
• Challenge: The Cold Brief. Timed live-problem session with an unseen brief, run in pairs (one performs, one observes with a rubric), then swap.
• Demo day: highlight reels + rubric scores.

• AI frees cognitive budget; the program directs the surplus at other humans. When everyone has access to the same intelligence, the machine side of every field flattens, and the human side becomes the whole game: trust, taste, teaching, recruiting, moving a room. The kids who invest their freed hours in humans compound in the one market AI cannot flood. Presenting is scary. But it is a skill, and it trains like one.
• The Last Bastion, tested honestly: pick a great you admire, name three strengths, test AI against them, write the thesis in the app's Human Edge builder. Some kids will conclude AI closes the gap in places: that is a legitimate, evidence-backed thesis. The skill is honest capability mapping, not a predetermined answer.
• Challenge: The Cold Ten. Ten real outreach messages to professionals in your lane (mentorship, feedback, or customers); response rate analyzed like a campaign. Plus one teach-down: teach a younger kid or peer one thing you mastered.
• Demo day: outreach results + ceiling thesis, pressure-tested by peers.

• Capstone ships in final form. Portfolio assembled: 3+ live creations, wrong-log, mission briefs, decision logs, outreach record.
• The Closed-Laptop Moment: presented live to a real audience (parents, mentors, and external practitioners, not just coaches): the problem chosen, why it mattered, how AI was leveraged, and where the magic was theirs alone. Laptop closed for the Q&A. Not a trial; a torch-pass. The kid pulls the room into the vision they have been building for twelve weeks.
• The "after" recording is cut against week 1's "before." That diff is the product.
• Graduation = the parent moment we design for: there is no way my kid made this; they are punching above their weight class.

“There is no way my kid made this. They are punching above their weight class.”

What gets measured is what stays scarce: judgment, ownership, and shipping.

• Closed-Laptop Defense, before vs after: ownership; the anti-slop mechanism.
• Ship cadence (weeks with a live deploy, W7-W12): bias to real; streak tracked in the crew.
• One-shot success rate: mission-brief quality; directing skill.
• Wrong-log depth: verification reflex; catches documented with sources.
• Problem-selection pitch score: the crown skill, graded by the mentors.
• Capstone presentation (external judges): the whole program, compressed.
• Humans moved: teach-downs given, demo days hosted, collaborators recruited, outreach responses.
• Ladder rung: Passenger to Typist to Director to Owner to Leader; where the kid sits, with the creations to prove it.
• Every creation carries the standard debrief. Every claim in a student's work needs provenance a human checked.

• Blown away. By week 12 you see work that makes you say "there is no way my kid made this."
• Creator screen time. Time in the program is building, presenting, shipping: creator time, not consumer time. The ROI on their future is visible in the portfolio.
• The kid yearns to log in. If they don't, we are doing it wrong.
• Inspectable, not magical. JARVIS is an AI system your kid can explain: model + memory + tools + mixing. No friend-fiction, no black boxes.
• Crutch-proof. This is the opposite of teaching kids to cheat. Every week ends in a laptop-closed explanation the machine cannot do for them. Kids who can't explain it didn't make it, and here, they always can.

• Magic prompt tricks. They expire with every model release. We teach the things that survive every model release: judgment, briefing, verification, taste, agency. Tricks expire; invariants compound.
• Coding syntax. We do not teach coding. We teach what makes the coding geniuses great: judgment and structured thinking. Picture it properly: your kid now has a fleet of Silicon Valley-grade software engineers standing by, ready to act the moment the mission is clear. The skill worth teaching is not typing what they type; it is directing what they build.
• Tool-of-the-month worship. Models rotate under JARVIS on purpose; skills anchor to components and systems, not product names.
• Output-graded work. Grading output in 2026 is grading the machine.
• Abstract AI ethics. Ethics shows up with teeth: real cases, real tradeoffs, argued from multiple positions, plus a personal strategy memo: "where I'll be irreplaceable in 2035 and why."
• Theory without shipping. Nothing in this program ends as a document nobody sees.

FutureMasters · The 12-week program for kids who will run the machines