Try this at Home: Jesse Genet on OpenClaw Agents for Homeschool & How to Live Your Best AI Life

What It Covers

Jesse Genet, former YC-backed startup CEO turned homeschooling parent of four, details how she built a five-agent OpenClaw system running on individual Mac minis to manage curriculum planning, content creation, finance, software development, and executive assistance — without any prior coding experience — transforming daily family logistics and personalized education delivery.

Key Questions Answered

• •Agent Specialization Over Generalism: Deploy purpose-built agents with single, narrow roles rather than one all-purpose assistant. Genet runs five separate OpenClaw instances — Sylvie for curriculum, Claire for chief-of-staff tasks, Cole for development, Theo for content, Finn for finance — each on dedicated Mac minis. Narrowly scoped agents maintain cleaner context windows, respond faster, and produce more consistent outputs than generalist agents juggling multiple domains simultaneously.
• •Employee Onboarding Framework for Agents: Treat AI agents exactly like new employees: provide role documentation, define communication protocols, grant minimal access on day one, and expand permissions only as trust develops. Start with read-only access to sensitive systems like email or calendar. Codify every workflow decision into markdown files the agent can reference persistently. This prevents agents from making autonomous decisions that violate unstated priorities or implicit boundaries.
• •Voice-First Delegation via Slack: Route all agent communication through Slack using voice notes as the primary input method. Agents transcribe voice memos and execute tasks without requiring the user to sit at a computer. Genet uses this to trigger curriculum generation, print documents to a physical printer, and coordinate multi-agent projects — all from a phone. This removes desk dependency and enables delegation during active parenting hours.
• •Curriculum Personalization at Near-Zero Cost: Use frontier models like Claude Opus to generate fully customized, 70-lesson annual curriculum progressions for specific children, incorporating Montessori methods, owned physical materials, and individual skill levels. Genet estimates full curriculum development costs approximately $8 in API tokens. Agents also analyze screen-recorded tutoring sessions via Loom transcripts to identify specific knowledge gaps — for example, flagging that a child confuses the numerals six and nine.
• •Multi-Agent Coordination via Shared File Protocols: Enable agents to collaborate by giving each a map of other agents' Slack bot IDs and channel IDs, stored in shared Obsidian markdown files. Designate one agent — in this case Claire — as a coordinator who sends task prompts to other agents via dedicated command channels, replacing generic heartbeat pings with context-aware instructions. This creates autonomous project handoffs where agents manage up and request approvals without constant human initiation.
• •Graduated Financial Access with Hard Limits: Issue agents a dedicated credit card with a low spending limit rather than granting access to primary financial accounts. This allows agents to autonomously purchase project supplies — such as e-ink displays for a home installation — while capping potential damage from errors. Pair financial access with logged purchase tracking in shared files so all transactions remain auditable and reversible within a defined risk threshold.
• •Local Hardware Integration for Physical-World Tasks: Connect OpenClaw agents to the local home network to give them access to physical devices including paper printers and 3D printers. Agents can execute print commands from voice instructions without any manual printer dialogue interaction. Local Mac mini hardware also offloads cron jobs from cloud inference, reduces token costs for routine heartbeat tasks, and supports privacy goals by keeping sensitive family data off third-party servers.