Proposal for foundation reviewPrepared by Jon Jessup
EAT. LEARN. PLAY.Family AI Coach
Build reading confidence and AI judgment through the experiences Oakland families already share.
Oakland children are already meeting machines that answer confidently and are sometimes wrong, usually with no adult beside them to show what to do about it. The same children are working to build reading confidence. Family AI Coach proposes treating those as one skill: a five-minute activity, run by a caregiver with a child aged 7 to 10, in which they read a short reviewed passage, answer a question, point at the sentence that proves it, and catch a mistake the Coach makes on purpose. Founded by Stephen and Ayesha Curry, EAT. LEARN. PLAY. supports Oakland children through nutritious meals, literacy resources, and equitable opportunities to play, and this proposal asks whether a small guided activity could travel with what those programs already put into families' hands.
What is new here is the deliberate mistake. Most family AI tools try to be right; this one is designed to be caught. Asking a child for the sentence that proves an answer is a reading exercise and a judgment exercise at the same moment, and it is the one move that stayed in the design from the first version to this one.
Who does what
The caregiver operates the service and the child supplies the ideas. A caregiver receives one short activity, reads it with their child, and asks the questions on the card. The child answers, points to evidence in the text, and decides whether an answer holds up. Nothing is designed for a child to use alone, and there are no child accounts. The initial age range is 7 to 10.
Start here
What is the Family AI Coach proposal?
Family AI Coach is a proposed service, not an approved foundation program and not a deployed Agentforce implementation. Every demonstration on this site is scripted and deterministic. Read the proposal one page at a time.
- The opportunityHow EAT, LEARN, and PLAY already reach Oakland families, and where a short guided activity could add reading confidence and AI judgment.
- Try the CoachA fully scripted demonstration of the family experience, including evidence checking, a deliberate mistake to catch, and an evaluation lab.
- Why AgentforceThe proposed technology: app, WhatsApp, and Apple Messages for Business channels, consent, routing, and output checks.
- Responsible designCaregiver guidance, no child accounts, minimal data, accessibility, consent, privacy, and the pause procedure.
- Pilot and impactA proposed 12-week pilot with targets, a 0–4 rubric, a blank budget calculator, and explicit decision steps.
- Evidence of iterationWhat changed along the way: concept versions tried and dropped, before-and-after Coach wording, and the questions that produced specific design changes.
- Share feedbackFive questions for reviewers. Answers are saved anonymously for proposal review.