// ai agent lab
Build your own AI agent.
A playful lab where anyone can build a robot helper, teach it with examples and test how smart it gets. No coding needed: just tap, train and try.
{{ agSay }}
skill: {{ agSkillLabel }}
brain power
{{ agAcc }}
examples learned: {{ agN }}
{{ agScore }}
color it
pick a skill
train it
test it
// how to play
Color it
Pick your robot’s color. Tap “new name” until you find one you like.
Pick a skill
Choose one job to teach: sorting recycling, cat or dog, or reading the mood.
Train it
Tap “feed an example”. Each one shows the right answer, like “banana peel → compost”.
Test it
Tap “ask”. Your robot sees something new and makes its best guess. Is it right?
// inside {{ agName }}’s brain
examples
training data
learning
training
model
what it learned
guess
prediction
what it learned so far
Each example adds to one answer. Lopsided examples make lopsided guesses.
// how it works
How your agent learns.
Examples
training dataYou show the agent examples with the right answer. Real AI learns the same way, from lots of labeled examples.
→
Spotting patterns
trainingEach example nudges the agent. After a few, it starts to notice what goes with what.
→
The model
what it learnedEverything it picked up is saved as a model: the agent’s “brain” for that one job.
→
Best guess
predictionAsk about something new and the model guesses, with how sure it is. Better examples, better guesses.
// brain power
Why more examples help.
In this lab, every example makes your robot smarter, but each new one helps a little less than the one before. That is why the bars grow fast at first, then slowly.
Real AI works in a similar way: more good, varied examples usually mean better guesses, up to a point. And even a well-trained AI can still be wrong sometimes, so people check its work.
brain power after this many examples
60%
1
67%
2
73%
3
78%
4
85%
6
90%
8
93%
10
96%
15
examples fed
// challenges
Try these challenges.
Speed trainer
Get your robot to 90% brain power. How many examples did it take?
Quiz master
Ask five questions in a row. Can your robot get all five right?
No training
Pick a skill and ask before feeding any examples. What does your robot say? Why?
New job
Teach one skill, then switch to another. Your robot starts fresh: each skill is a new model.
// ai around you
AI you already use every day.
Spam filters
Learn from emails people mark as junk, then move new junk out of your inbox.
Photo apps
Learn from labeled pictures so they can find pets, faces and places in your photos.
Recycling robots
Use cameras and AI to sort plastic, paper and metal on fast conveyor belts.
Music and video picks
Learn what you like from what you play and skip, then suggest something new.
// words to know
Speak AI.
# AI agent
A program that takes in information, decides and acts to get a job done.
# Training data
The examples, with the right answers, that an AI learns from.
# Model
What the AI learned, saved so it can make guesses later.
# Prediction
The AI’s best guess about something it has not seen before.
# Accuracy
How often the guesses are right. Our robot calls it brain power.
# Bias
When the examples are lopsided, the guesses come out lopsided too.
$ kids-in-ai book --tech-day
Bring the AI Lab to your school.
> At a Kids IN AI Tech Day, students train real AI models, like object classifiers in Teachable Machine, with our team.
