The learner builds
They make the important choices; AI does not complete the entire task for them.
AI learning games
A game should make AI thinkable—not magical.
Wovi combines the motivation of an adventure with the discipline of a good experiment: form an idea, test it, see the evidence and try again.
FUTURE MISSION CONCEPTSKY CITY
Clear the airways
Teach three courier agents to share routes without causing a traffic spiral.
Concept: planning + coordination
The productive struggle
Four signs of a meaningful AI learning game
Behavior is visible
They can inspect why an agent acted and compare that with the intended goal.
Failure is useful
A weak result becomes evidence for the next experiment, not a dead end.
Reflection is part of play
The mission asks the learner to explain what changed and why it mattered.
The Wovi loop
Adventure on the surface.
Scientific thinking underneath.
1DiscoverA world problem
→2DesignAn agent strategy
→3DeployInside the simulation
→4DebriefEvidence and tradeoffs
WORLD 01 · PLAYABLEReef RescuePatterns & classification
WORLD 02 · PLAYABLEAgent MazeRewards & Python
WORLD 03 · OPTIONAL PLAYTESTSignal ForestData quality & bias
Scenario themes
New world.
New way to think.
Each world introduces a different relationship between people, intelligent agents and the environment. Concepts return in new contexts so understanding can transfer.
Agent Maze and Reef Rescue are the two released prototypes available from Atlas Station. Signal Forest offers an optional early playtest with no saved progress; Sky City remains a future concept.
Play Agent Maze →Founding families
Want to test the first AI learning missions?
Join the early-access list for product previews, parent research invitations and launch updates.
Join early access
Free to join · Parent email only · Unsubscribe anytime