Education

How Jesse Genet Uses 5 OpenClaw Agents for Homeschool and Inventory

Built by Jesse · · Source: the web

Quick answer

Jesse Genet runs five OpenClaw agents, each on its own Mac Mini. Sylvie turns photographed curriculum books into lesson guides and catalogs the family's learning materials in Obsidian, Cole built a kids' TV app called Mira, and Finn handles finance with read-only access.

Result: Curriculum photos became lesson guides and printable illustrations, the Mira kids' TV app was built in four days and runs on a Google TV streamer, and a photographed worksheet can be printed in about 30 seconds.

What they built

Jesse Genet is a homeschooling parent and entrepreneur who runs five OpenClaw agents, each on its own Mac Mini. The setup is described in a How I AI episode write-up by Claire Vo, which names three of the five agents:

  • Sylvie handles homeschool work. It reads source material, maintains an Obsidian vault, writes lesson guides, makes illustrations, catalogs physical inventory and is connected to a printer.
  • Cole handles coding. It built Mira, a kids' TV app, and found a way to run it on the family TV.
  • Finn handles finance. It has read-only access to financial information and only communicates through Slack.

The write-up doesn't describe the other two agents, and it gives no detail on the finance workflow beyond Finn's access.

How it works

Lesson planning. Jesse photographed curriculum books, including Teach Your Child to Read in 100 Easy Lessons and a BFSU science curriculum. According to the write-up, Sylvie used the photos to record each book's structure, concepts and lesson order in Obsidian, then wrote a one-page teaching guide per chapter with objectives, vocabulary and required materials. For a lesson on animal adaptations and survival, Jesse asked for child-friendly watercolor illustrations sized for 8.5 x 11 paper, and Sylvie generated them with a Google Gemini model. The notes live in Obsidian so Jesse can correct them.

Physical inventory. Jesse sends Sylvie photos of books, toys and Montessori materials. Sylvie creates an Obsidian entry for each item with its name, type, estimated age range, description, tags and links to related lesson plans. Because inventory and lessons share one system, Sylvie can suggest materials the family already owns. Jesse can also photograph a worksheet and tell Sylvie to print it.

The Mira app. Jesse wanted TV time limited to approved YouTube channels, with no comments, recommendations or unrelated videos. Mira builds themed streams from those channels, and the kids' controls are limited to go, forward, back and pause. Jesse built it by describing features and testing what Cole produced, often from a phone in short sessions. When Jesse asked how to get it on the family TV, Cole first said it wasn't possible. Jesse pushed back, and Cole found a route through a Google TV streamer and walked Jesse through packaging and deploying it.

Access rules. Each agent has one job, a bounded set of data and a set communication channel, which the write-up says makes errors easier to trace. No agent has full read and write access to Jesse's email, and permissions are added over time. Roles and personas live in soul.md files. When Jesse labels something a decision, the agent records it in a decision file so later sessions don't reopen it. Jesse treats setup like hiring an employee and still picks the curriculum, approves channels, tests the app and grants permissions.

The result

According to the write-up, curriculum photos became structured lesson guides and printable materials, and shelf photos became searchable inventory entries that feed into lesson planning. Mira took four days to build and runs on a Google TV device, even though Jesse had first opened a terminal only six months before the episode. A photographed worksheet can be printed in about 30 seconds.

How to build something similar

  • Start with one agent and one recurring chore that has a clear input and output, such as photographing a single shelf or indexing one chapter of a book. You don't need five machines to try this.
  • Give the agent only the access that job needs. Start with read-only where you can and add permissions after you've watched it work.
  • Keep its output in plain notes you can open and edit, like an Obsidian vault, so you can fix mistakes and the agent can use your corrections next time.
  • Write down decisions in a file the agent reads each session so settled questions stay settled.

The original post

Claire Vo · View on the web

Want to build something like this?

Join the Shipping Skool community free. Paid members get live bootcamp calls every week.

Join Shipping Skool — Free

More use cases

← All AI agent use cases