Set one learning objective, one small deliverable and one explicit boundary on release. The following is a proposed six-session plan, not a promise that every teenager can complete the same project in six weeks.
Explore the comparison in this briefing
Session
1. Scope and baseline
- Main work
- Define the user, limits, starting skills and permitted tools
- Evidence to retain
- One-page brief and a small starting task
Compare each option against the same criteria. The complete table and qualifications remain in the briefing.
Illustrative brief: build a private FAQ prototype using synthetic information, with no public accounts, payments or sensitive data.
| Session | Main work | Evidence to retain |
|---|---|---|
| 1. Scope and baseline | Define the user, limits, starting skills and permitted tools | One-page brief and a small starting task |
| 2. Simplest baseline | Build a basic lookup or rule-based version | Learner explains what works without generative AI |
| 3. Add the AI component | Introduce an approved model connection if appropriate | Diagram of inputs, processing and outputs |
| 4. Evaluate | Try normal questions, unsupported questions and misleading inputs | Short test log, including failures |
| 5. Improve and transfer | Make one justified change; tackle a related unfamiliar task | Explanation of the change and assistance required |
| 6. Demonstrate and hand over | Present limitations, contributions and next steps | Runnable files, instructions and contribution record |
Write the assumed between-session time into the agreement. If the teenager has no time outside lessons, reduce scope rather than silently increasing mentor production.
Treat any external-action capability separately. OWASP identifies excessive permissions and autonomy as sources of risk in AI applications; our default for a first project is no email sending, purchasing or access to household systems. (genai.owasp.org)