Descriptions:
In this interview on the How I AI podcast, host Claire Vo speaks with Jesse Genet — entrepreneur, homeschooling parent of four, and early power user of OpenClaw (Claude-based local agents) — about how she runs five separate AI agent instances to manage her household, finances, homeschooling curriculum, and communications.
Genet’s setup involves five Mac minis, each running a distinct OpenClaw instance with strictly scoped file system access and communication permissions. One agent, nicknamed Finn, has read access to bank account statements but no outbound communication channels at all, preventing any accidental data leakage. Another agent, Claire, can send iMessages for scheduling but deliberately cannot access financial documents. This physical and logical partitioning mirrors how Genet thinks about onboarding employees — building trust incrementally rather than granting blanket access from day one.
The conversation surfaces practical security thinking that is rarely discussed in mainstream AI agent coverage: the risks of over-permissioning, the difference between an agent that impersonates you versus one that assists you, and how Obsidian markdown files serve as the shared knowledge layer connecting agents to her personal context. For anyone exploring multi-agent home or personal productivity setups, this episode offers one of the most concrete real-world architectures available, grounded in a non-technical founder’s lived experience rather than developer demos.
📺 Source: How I AI · Published February 25, 2026
🏷️ Format: Interview
