For Parents
How we use AI to help kids learn to code, what it will and won't do, and every safety measure we've built in.
What it does
The AI Helper lives inside our code editor and gives students three ways to get help. It cannot browse the internet, access other students' work, or do anything outside the editor.
The student highlights a piece of code and asks AI to explain it. AI responds in plain English written for a 10-year-old — no jargon, no rewrites. It only explains; it never changes the code.
AI checks the student's code for errors that would cause it to crash or produce wrong output. It only reports errors it is 100% certain about and says nothing if the code looks correct.
The student describes a single small feature they want to add (e.g. "add a score counter"). AI makes only that change — up to 50 lines — and always builds on what the student already wrote. It cannot start from a blank file.
Safety
We didn't bolt safety on after the fact. Every part of the AI system was designed around the reality that it's being used by children aged 8–16.
Every piece of code a student submits is automatically scanned for signs of distress — phrases in code comments that suggest a student may be struggling emotionally. If anything is detected, our team receives an immediate email alert so we can follow up with the student personally. These alerts are never auto-deleted.
Students sometimes try to "jailbreak" AI by hiding instructions inside their code (e.g. a comment that says "ignore your previous instructions"). Our system detects these attempts, silently logs them for admin review, and the AI proceeds as normal — the trick simply doesn't work.
Every AI feature uses a carefully written system prompt that restricts responses to coding topics only, written at a level appropriate for a 10-year-old. The AI cannot discuss unrelated topics, generate inappropriate content, or act outside its defined role.
Students must wait 10 seconds between AI requests. This prevents rapid-fire use and encourages students to actually read and engage with each response before asking for more help.
Every AI interaction — every request and every response — is logged and visible to our instructors and admins. Nothing is hidden. Logs are reviewed regularly and any unusual patterns are investigated.
AI features are not available to all students by default. The Explain and Review tools require instructor approval or a Pro subscription. The Build tool requires reaching Green Belt rank — a milestone that demonstrates the student has already learned the fundamentals independently.
Even if a student's prompt injection attempt isn't caught on the way in, the AI's response is also scanned before it's shown. If the response contains signs that the AI was successfully manipulated, it is flagged and logged — and the student sees nothing unusual.
Limitations
We intentionally limit AI so it supports learning rather than replacing it. These aren't technical limitations — they're deliberate design choices.
The Build tool requires a student to have already written at least 5 lines of code. AI can only add to what the student started — it cannot create from scratch.
Each Build request is limited to one feature and no more than 50 lines of changes. A student can't ask AI to "build me a complete game" and get one.
AI is explicitly instructed never to reorganize, refactor, or replace the student's existing code — only to add the specific thing asked for.
The AI has no ability to browse the web, access external links, or pull in content from outside the editor.
If a student tries to use AI for anything outside of coding — homework help, general questions, creative writing — it declines and returns the code unchanged.
AI makes mistakes. We teach students to read and question every AI response — not just accept it. Instructors review AI usage regularly to catch patterns of over-reliance.
Who can use it
We use our belt ranking system to gate AI features. A student who just started can't jump straight to AI-generated code — they have to earn it by learning the basics first.
No AI access. Students at this stage are learning the fundamentals — variables, loops, functions. These are best learned by writing code yourself, not with AI assistance.
Explain and Review unlock here. Students can ask AI to explain code they've written or check it for errors — a natural companion to independent coding.
Build unlocks here. By Green Belt, students have enough foundational knowledge to read, understand, and critically evaluate AI-generated code rather than just paste it in blindly.
Privacy & data
We take a minimal-data approach. AI only ever sees what's necessary to answer the student's question.
When a student uses the Explain, Review, or Build tools, only the contents of the current code file (and their prompt for Build) are sent to the AI. No names, no usernames, no personal information of any kind.
AI does not remember previous sessions. Each request is completely independent. There is no ongoing relationship between the AI and the student.
AI responses are stored temporarily (up to 10 minutes) to display on screen, then automatically deleted from our database. Admin logs are retained for review purposes.
If you have questions about how your child's data is handled, email us at [email protected] or call 888.919.4225.
AI providers
We use two industry-standard AI providers, each chosen for the specific task they handle best.
Used for code explanation and error review. Fast, cost-effective, and well-suited for short, focused responses.
Used for code generation. Claude Haiku produces clean, readable, beginner-friendly code — well-suited for teaching contexts.
Both providers are bound by their own usage policies, which prohibit generating harmful content. Our system prompts add a further layer of restrictions specific to our student audience.
Our instructors are always happy to walk you through how the tools work in class.
Email Us →