How I AI: Jesse Genet’s 5 OpenClaw Agents for Homeschooling, App Building, and Physical Inventories

In this episode of How I AI, homeschooling parent and entrepreneur Jesse Genet shows how her OpenClaw agents turn photographed curriculum into lessons, build a curated TV app, and connect a physical inventory of books and toys to daily planning.

Jesse runs five specialized agents on separate Mac Minis. She gives each one a role, limited access, and its own context instead of asking one agent to manage the entire household.

The three demonstrated workflows use the same pattern: put real household context into the system, give one agent a bounded responsibility, and reduce the number of manual steps between a request and a usable result.

Turn curriculum photos into tomorrow's lesson

The homeschool planning workflow connects curriculum capture, lesson generation, printable materials, and progress notes without requiring a parent to retype the books.

The quality of the output depends on the quality of the source. Jesse gives Sylvie the actual curriculum sequence and keeps the resulting notes in Obsidian, where she can correct them. The agent is preparing from the family's materials, not inventing a generic school day from scratch.

Jesse kept homeschool notes in Obsidian but did not have time to enter every lesson, resource, and child-progress update by hand.

She added a homeschool agent, Sylvie, on top of the Obsidian vault to read source material and maintain the files.

Photograph the source material

Jesse photographed curriculum books, including Teach Your Child to Read in 100 Easy Lessons and the BFSU science curriculum, instead of transcribing them.

Sylvie used the images to recover the books' structure, concepts, and lesson sequence for the Obsidian knowledge base.

Generate the plan and printable materials

Sylvie turns a chapter into a one-page teaching guide with objectives, vocabulary, and the materials required for its activities.

For an animal-survival lesson, Jesse also asked for printable illustrations:

I want watercolor illustrations suitable for kids that can print on eight and a half by 11 of each of these concepts.

Sylvie used a Google Gemini model to generate watercolor illustrations for camouflage, defense, and finding food. Jesse thinks the creative teacher persona in Sylvie's soul.md file may have influenced the style, but she did not present that as a proven cause.

Build the exact kids' TV experience you want

The slop-free kids TV workflow follows Mira from the approved channel list through the deliberately tiny interface and onto the Google TV device.

The constraint is the product. Mira does not need comments, discovery, or an engagement loop. It needs to play from a parent-approved set and offer controls a child can understand. Removing features is what makes it useful.

Jesse wanted a viewing experience made only from channels she had chosen, without comments, recommendations, or unrelated videos.

She worked with her coding agent, Cole, despite having opened a terminal for the first time only six months earlier.

Build from the parent's requirements

The app creates themed streams from approved YouTube channels. Its child-facing controls are limited to Go, forward, back, and pause.

Jesse built the app over four days, often from her phone in short sessions, by describing features and testing Cole's implementation.

Carry the prototype all the way to the TV

When Jesse asked how to run the app on the family TV, Cole initially said it was not possible. She pushed back:

Try harder, Cole. Okay, that's not an answer we need right now. We've got real work to do, guys. Save these kids' souls, Cole.

Cole found a route through a Google TV streamer and guided Jesse through packaging and deploying Mira to the device.

Make the contents of the cupboards searchable

The physical inventory workflow shows how to turn shelf photos into structured entries the planning agent can retrieve later.

This closes an easy-to-miss context gap. A language model may know thousands of activities, but that knowledge is less useful than knowing which book, puzzle, or Montessori material is already ten feet away.

An agent cannot recommend a physical book or toy if it does not know the item exists. Jesse addressed that gap by photographing the learning materials in her house.

Books and educational supplies were sitting unused in cupboards because Jesse often remembered them after a child's interest had passed.

Photograph and catalog what you own

Jesse sent photos of books, toys, and Montessori materials to Sylvie with this request:

I wanna make an inventory of my learning supplies. Here's the photos.

Sylvie created structured Obsidian entries with the item, type, estimated age range, and description.

Put the inventory inside the planning loop

Because the inventory and lesson plans share a system, Sylvie can suggest a material Jesse already owns when preparing a lesson.

Jesse also connected Sylvie to a printer. Instead of scanning, emailing, downloading, and printing a worksheet, she can photograph it and say:

Sylvie, print this.

The worksheet can be ready in about 30 seconds, which removes several device handoffs from the middle of a lesson.

How Jesse limits and manages the agents

The five-machine setup is Jesse's implementation, not a requirement for every household. The transferable design is separation: each agent has a job, a bounded data surface, and a channel where its actions can be observed.

Jesse expands access only when the work justifies it. Read-only finance access can support analysis without allowing transactions. A homeschool agent can edit the Obsidian vault without reading email. A coding agent can work on the app without inheriting every piece of family context.

That division also makes failures easier to diagnose. When an output is wrong, Jesse knows which role, source material, and permission set produced it. One all-purpose household agent would be harder to understand and harder to trust.

Jesse treats agent setup like onboarding an employee: assign a job, limit access, add context, and expand trust only after observing the work.

• Separate machines and channels: Jesse runs each agent on its own Mac Mini. Finn has read-only access to financial information and communicates only through Slack. Other agents have different tools and data based on their jobs.

• Progressive access: None of Jesse's agents has full read-write access to her email. One agent has read-only access. She adds permissions gradually instead of granting broad access at setup.
• Specialized roles: Sylvie handles homeschool work, Cole handles coding, and Finn handles finance. Each soul.md file describes the role and persona.
• Decision file: When Jesse says that something is a decision, the agent records it in a decision file so later sessions do not reopen the same question.

The useful constraint is not the number of agents. It is that each one has a defined job, a limited set of data, and an explicit communication route.

The memorable part of Jesse's system is not that agents live on Mac Minis with names. It is that the digital work connects back to physical family life: a curriculum book becomes a lesson, a shelf becomes a searchable inventory, a worksheet reaches the printer, and a narrow app reaches the television.

Each workflow removes device handoffs and administrative steps while leaving the parent in charge of the choices. Jesse selects the curriculum, approves the channels, tests the app, and decides which permissions an agent earns.

A useful starting point is one recurring bottleneck with a clear source and destination. Photograph one shelf, index one curriculum chapter, or curate one channel list. Give the agent only the access needed for that job, inspect the result, and record corrections where the next run can use them.