Test teaching behaviour and relevant technical judgement separately. A polished demonstration answers neither question adequately.
Explore the comparison in this briefing
Time
5 minutes
- Activity
- Learner interest and goal
- What the parent should notice
- Does the mentor listen rather than deliver a sales pitch?
Compare each option against the same criteria. The complete table and qualifications remain in the briefing.
Even the job title can be ambiguous. One reviewed employer advertisement for an “AI Instructor for Kids” emphasised using AI tools and working with children, while coding experience appeared as a desirable rather than essential qualification. That is not evidence against the role; it shows why an AI-instructor title alone does not establish machine-learning engineering depth. (indeed.com)
What should you ask in the adult interview?
Use an approximately 30-minute conversation:
- Starting point: “What would you check before accepting this project?”
- Scope: “What is the smallest version the child could genuinely own?”
- Teaching: “How would you help without taking over when they are stuck?”
- Technical judgement: “Show how you would diagnose an unfamiliar small bug or an implausibly strong model result.”
- AI assistance: “Which parts may AI generate, and which parts must the learner practise?”
- Evidence: “What would your progress report show beyond completed features?”
- Boundaries: “What happens if the child messages privately or enters personal information?”
- Continuity: “Who takes over if you leave, and what is transferred?”
Ask for a paid assessment if the technical exercise needs substantial preparation or specialist time. Do not request free engineering work disguised as an interview.
What should a 50-minute trial look like?
Agree a modest task in advance and tell everyone what will be observed.
| Time | Activity | What the parent should notice |
|---|---|---|
| 5 minutes | Learner interest and goal | Does the mentor listen rather than deliver a sales pitch? |
| 10 minutes | Explain or trace a small example | Does the mentor discover what the child understands? |
| 20 minutes | Guided building or debugging | Who is making decisions and operating the workspace? |
| 10 minutes | Small changed requirement | Can the learner adapt the work with appropriate support? |
| 5 minutes | Learner recap and next step | Can the child describe the lesson in their own words? |
Keep any parent observation disclosed. Afterwards, ask the child privately: “Did you feel comfortable asking questions?”, “What did you actually figure out?” and “Would you like another session with this person?”
How should you score candidates?
First record the non-negotiable gates: identity and screening process, communication boundaries, lawful and eligible tool use, data arrangements and willingness to respect the child.
For candidates who clear those gates, use this proposed comparison rubric:
| Criterion | Weight | Evidence |
|---|---|---|
| Teaching and response to confusion | 30% | Observed explanation, questioning and adaptation |
| Relevant technical judgement | 25% | Live reasoning, debugging and evaluation choices |
| Learner rapport and agency | 20% | Child feedback and willingness to participate |
| Realistic scope and learning evidence | 15% | A small project with clear independent capabilities |
| Reliability and continuity | 10% | Scheduling, references and handover arrangements |
Rate each criterion from 0 to 3 and calculate weight × rating ÷ 3. Record “not established” separately when evidence is missing. The total helps organise a decision; it is not a safeguarding certification or a scientifically validated predictor.
Pause or reject a proposed arrangement if the mentor requests false-age accounts, concealed communication, unnecessary sensitive data, substantial unapproved work or a project they expect to complete mainly themselves. A learner’s continuing discomfort is also sufficient reason not to proceed.