5 OpenClaw agents run my home, finances, and code | Jesse Genet

February 25, 2026

49 min episode · 2 min read

Jesse Genet

Episode

49 min

Read time

2 min

AI-Generated Summary

Published Feb 25, 2026

Key Takeaways

✓Agent Partitioning by Role: Assign each OpenClaw agent a single domain — homeschool, finance, coding, scheduling — and give it access only to the relevant file vault. Jesse runs five agents on separate Mac minis specifically to prevent sensitive financial data from leaking into agents with outbound communication access like iMessage or Slack.
✓Photo-to-Structured-Data Workflow: Photograph physical books, worksheets, and educational materials, then instruct the agent to generate structured lesson plans, inventories, or printable materials from those images. Jesse photographed an entire science curriculum book and had her agent produce formatted lesson plans with objectives, vocabulary, materials lists, and suggested activities for each chapter.
✓Physical Inventory Linking to Curriculum: Photograph all educational toys, books, and supplies once to create an agent-readable inventory. The agent then cross-references owned materials against existing lesson plans and proactively suggests which physical items to retrieve from storage when planning a specific lesson — eliminating the problem of rediscovering resources at the wrong developmental stage.
✓Progressive Trust Onboarding for Agents: Treat agent setup like onboarding a new employee — start with read-only access, never grant impersonation rights over primary email, create a separate email address and dedicated password vault for the agent. Jesse gave her scheduling agent calendar delegation access and its own credentials rather than sharing her personal login.
✓Soul File and Decision File Architecture: Maintain a `soul.md` file defining each agent's persona and a separate decisions file logging final, non-reversible choices. When behavior drifts, instruct the agent to self-diagnose against its soul file and propose edits rather than manually rewriting it. Saying "that's a decision" signals the agent to log it and stop re-raising the topic.

What It Covers

Jesse Genet, mother of four and homeschool parent, runs five separate OpenClaw agents on dedicated Mac minis to manage homeschooling curriculum, personal finances via QuickBooks, coding projects, scheduling, and household inventory — demonstrating how AI agents can restore productive capacity to time-constrained parents without requiring technical expertise.

Key Questions Answered

•Agent Partitioning by Role: Assign each OpenClaw agent a single domain — homeschool, finance, coding, scheduling — and give it access only to the relevant file vault. Jesse runs five agents on separate Mac minis specifically to prevent sensitive financial data from leaking into agents with outbound communication access like iMessage or Slack.
•Photo-to-Structured-Data Workflow: Photograph physical books, worksheets, and educational materials, then instruct the agent to generate structured lesson plans, inventories, or printable materials from those images. Jesse photographed an entire science curriculum book and had her agent produce formatted lesson plans with objectives, vocabulary, materials lists, and suggested activities for each chapter.
•Physical Inventory Linking to Curriculum: Photograph all educational toys, books, and supplies once to create an agent-readable inventory. The agent then cross-references owned materials against existing lesson plans and proactively suggests which physical items to retrieve from storage when planning a specific lesson — eliminating the problem of rediscovering resources at the wrong developmental stage.
•Progressive Trust Onboarding for Agents: Treat agent setup like onboarding a new employee — start with read-only access, never grant impersonation rights over primary email, create a separate email address and dedicated password vault for the agent. Jesse gave her scheduling agent calendar delegation access and its own credentials rather than sharing her personal login.
•Soul File and Decision File Architecture: Maintain a `soul.md` file defining each agent's persona and a separate decisions file logging final, non-reversible choices. When behavior drifts, instruct the agent to self-diagnose against its soul file and propose edits rather than manually rewriting it. Saying "that's a decision" signals the agent to log it and stop re-raising the topic.

Notable Moment

Jesse built a functional curated YouTube streaming app for her children — with forward, back, and pause controls only — and extended it to a physical Google TV Streamer device, all within four days of evening coding sessions conducted via phone while managing four young children.

Know someone who'd find this useful?