Observe the world
The scout sees valid nearby moves, distance to the beacon, energy cells, storm tiles and visited locations.
AI agents for kids · Ages 9–13
Give an agent a goal.
Debug the rules behind its choices.
An AI agent should not feel like an invisible mind. In this free activity, children predict a route, inspect a reward-scoring failure and change the safeguard that maps directly to Python.
AGENT LOOPA weak safety penalty makes the dangerous shortcut look valuable.
Observe → score → act → inspect
+4 move closer
+2 collect energy
−2 enter a storm
BEST FORAges 9–13Curious beginners
TIMEAbout 5–8 minutesOne complete debug loop
CODEBlocks beside PythonNo prior coding needed
ACCESSFree in a browserNo child account
A practical definition
What is an AI agent
for a young learner?
An agent observes a bounded environment, scores available actions and chooses one toward a goal. People decide what it can see, what earns reward and what must be protected.
Score each action
Progress, energy, safety and memory each contribute a visible number rather than a hidden explanation.
Choose and act
The agent selects the highest-scoring valid move. It follows the rule exactly—even when the human intention was incomplete.
Debug with evidence
The learner compares a prediction with the route, strengthens the storm penalty and reruns the same world.
TERM CHECK · REINFORCEMENT LEARNING
A reward score does not mean the agent is learning.
Agent Maze uses a fixed policy: the learner changes a number and reruns it. Reinforcement learning would update a policy from repeated experience and feedback. This mission teaches the goal-and-reward foundation without mislabeling the algorithm.
Free AI agent game
Agent Maze turns failure
into a debugging clue.
The starter policy offers progress and energy rewards but underprices a storm. The resulting collision makes the missing safeguard visible.
- Predict before running the policy
- Watch every move in a deterministic grid
- Change one reward value at a time
- Read the equivalent Python class and conditions
VISUAL RULE → PYTHON
Strengthen safety
STORM_PENALTY = 12
if move.is_storm:
value -= STORM_PENALTYA wider learning path
Build the mental model,
then grow the code
Agent Maze is the first bridge. Learners can next explore longer Python projects, compare agent behavior with classification, and carry both capabilities into connected worlds.
AI agents, clearly explained
Questions from families and beginners
What is an AI agent for kids?
An AI agent is a system that observes a situation, scores possible actions and chooses an action toward a goal. For children, the important idea is that the goal, available observations and safeguards are designed by people.
Can kids build an AI agent without coding?
Yes. A visual activity can expose goals, rewards and safety rules first. Wovi then shows the matching Python so the learner can connect a visible rule with variables, conditions and a scoring function.
How is an AI agent different from a chatbot?
A chatbot mainly responds in conversation. An agent takes actions inside an environment. Agent Maze is a bounded simulation with inspectable rules, not an open-ended child chatbot.
What does Agent Maze teach?
The mission asks learners to predict a route, observe an unsafe shortcut, inspect why the scoring rule caused it, strengthen a safeguard and rerun the same world.
Is Agent Maze reinforcement learning?
No. The current mission uses a fixed, transparent reward-scoring policy and does not learn a policy through repeated experience. It introduces goals, rewards and evaluation accurately as foundations that can support a later reinforcement-learning lesson.
Is an account required?
No. The current activity runs free in a browser and does not ask for a child's name, email, birthday, school or open-ended profile.
Founding families
Want more build-and-debug agent missions?
Join the adult early-access list for new playable worlds, curriculum updates and parent research invitations.
Join early access
Free to join · Parent email only · Unsubscribe anytime