Every other robot is a black box. GlassBox makes it glass. Kids build real agentic AI — Listen, Reason, Act — on a Wonder Workshop Dash robot, and watch the reasoning happen live.
Most robot kits teach kids to give commands: press play, robot moves. GlassBox teaches kids to build an agent — drag-and-drop blocks that let a real Dash robot listen, reason, and act on its own, while every decision shows up on screen as a visible reasoning trace. Kids don't just watch the robot move. They watch it think.
GlassBox runs on-device — no cloud costs, no subscription required to try it, and it works with the Wonder Workshop Dash hardware your program may already own. Small groups, hands-on, ages 8–12. A curious kid and a robot that explains itself, side by side.
Drag-and-drop agent blocks — Listen, Reason, Act — that control a real Wonder Workshop Dash robot over Bluetooth. No cloud, no wires, no assembly required.
To read a visible reasoning trace and understand why an AI agent chose an action — not just that it moved. The same ReAct pattern used in real-world systems, made visible and kid-sized.
Runs on-device, model-agnostic, and pairs directly with Dash hardware already in classrooms and after-school programs. No new robots to buy.
GlassBox isn't a single lesson — it's an 8-week arc that takes kids from “press play” to genuinely understanding how an AI agent reasons, remembers, and sometimes fails.
The starting point every coding robot shares: simple step-by-step commands. Kids build a “robot parade” and see exactly how a sequence of instructions becomes real movement.
Kids introduce branching logic: the robot senses its surroundings (an obstacle, a sound, a touch) and picks between two possible actions. This is the moment a robot stops just executing and starts deciding.
The hero week. Kids describe a goal in plain language, and the agent breaks it into steps on its own — Make List, One Step, Do It, Check Off. This is a real ReAct loop, the same pattern used in modern AI agents, just made visible and playable.
Kids give their agent memory across a session — and see, live, what happens when that memory fails or gets confused. The failure is the lesson: agents aren't perfect, and understanding why they get things wrong is real AI literacy.
The pilot-anchor week. Kids predict what the agent will do before it moves, based on reading its trace log. This is the single clearest moment where “black box” becomes “glass box.”
Kids learn to pause a running agent, inspect what it was about to do, and change course — the beginning of real debugging skill, not just trial and error.
Kids see firsthand what on-device AI can and can't do: hallucination, sensor confusion, and how to read a trace to figure out what went wrong. This is “physical AI engineering,” not magic.
A self-directed final project. Kids design their own agent from scratch and demo it — often the moment parents see the whole arc come together.
Every week maps directly to real agentic AI concepts — the same ideas behind modern AI systems — translated into blocks an 8-year-old can drag, drop, and understand.
Every decision the robot makes shows up as a visible trace — not hidden inside an app, not a mystery. Kids see the why, not just the what.
Runs on the Dash robot your program already has. No new equipment, no new budget line.
Small groups, hands-on, ages 8–12. Not a lecture — a robot and a curious kid, working it out together.
Starts with simple step-by-step commands and ends with genuine agentic reasoning — an 8-week arc, not a single afternoon.
The same reasoning pattern — Listen → Reason → Act — used in real-world AI agents today, made visible, playable, and kid-sized.
Demo video in production. Until then, the canvas above is the show: kids watch the robot’s reasoning happen live.
GlassBox started as a handful of pilot workshops, and we're growing it one program at a time. If you run a coding school, an after-school program, or a library STEM night — or if you're a parent who wants your kid to try it — we'd love to talk.
