Resource
AI Guidance by Age
Most AI guidance for education is written for one age or one role, and quietly assumes a generic learner. This does the opposite: it fixes an age band and a developmental capacity, then reasons about that specific pairing. The premise is that the same tool meets a very different mind at five, fifteen and thirty-five, so advice that ignores which capacity is still forming is advice about the tool rather than about the learner.
What you can do with it
Pick a role (caregiver, K-12 or higher-education educator, administrator, workforce educator, student, researcher or policymaker), then a developmental domain and an age band, and the guide returns one entry. The domains run from cognitive foundations, knowledge and understanding, and reasoning and metacognition through language and literacy, numeracy, creativity, agency and motivation, emotion, social understanding and interaction, to relationships. The bands run from 0–2 to 26–40.
Each entry has the same shape: a quick take, a numbered set of recommendations, the developmental processes and milestones that research already establishes for that age, then the two speculative sections on how AI might hinder and how AI might help, followed by further reading. A worked example at 15 to 17 for reasoning and metacognition recommends asking the teen for a first attempt and a stated confidence level before allowing AI, then having them use AI feedback on their own draft to find missing evidence or weak warrants. Entries can be downloaded as a PDF.
Who it is for
Early childhood and K–12 educators deciding what to allow at which age; parents and caregivers looking for a defensible line rather than a rule of thumb; administrators and policymakers who need the developmental argument in one place. Researchers will find it useful as a map of where the hypotheses are thin.
Notes and caveats
The guidance is explicit about its own status. Its authors call the entries hypotheses built by abductive reasoning from established developmental science plus early, often small, AI research. It is not conclusive guidance, because systematic evidence on how AI shapes a developing mind may take a decade. The tool models careful deliberation rather than reporting effects.
Two further limits matter for teaching. The recommendations are design prompts, not tested interventions; nothing on the site reports an outcome. And the age-band structure imposes a coarse grid on development that varies within bands, across cultures and across contexts, which the guide itself acknowledges.
The framing worth carrying into practice is the site's own: keep the loop human, then work out where AI fits inside it, rather than putting a human checkpoint inside a loop that belongs to the machine. That is the same concern the research records as offloading risk when a tool takes over the part of the task that was doing the developmental work.