AI Lab
Challenge
Three young kids, a dog, two working parents, and a firehose of school emails, activity signups, meal decisions, and calendar collisions. Running a household is a full-time operations job. Inspired by Rose Yao's family-assistant playbook, I wanted to test whether an AI agent could actually carry some of the mental load. Not simply answer questions about the household, but own its inbox, docs, and schedules — with boundaries I'd trust.
Strategy
I gave the assistant its own identity: a dedicated Google account with its own Gmail, Calendar, and Drive, fed by auto-forward filters that route school and activity mail out of our inboxes and into its own. Then a knowledge layer it maintains itself — our family handbook plus companion docs covering meals, weekly rhythm and protected time, and caregiver handoff and emergency info. Standing instructions define the autonomy contract: it reads and drafts everything freely, builds Instacart carts from the week's meal plan, and keeps the calendar clean. Scheduled runs close the loop: a daily 7 AM brief, a Friday weekend preview, Sunday planning prep — each refreshing a visual dashboard at a stable URL and emailing me the summary.
Results
The entire build took about a week of evenings. School email now lands sorted in the assistant's inbox. Recipe links become structured meal docs become a ready-to-checkout grocery cart in one pass. The real design work turned out to be the trust architecture — deciding exactly what the agent owns, what it drafts, and what stays human. Get that right, and the mental load starts to move.