This Mom Runs Her Family on 5 AI Agents. No Coding Required.

Jesse Genet is a homeschooling mom of four. Six months ago she had never opened a terminal. Today she runs 5 AI agents on 5 Mac Minis. One plans her lessons. One builds custom apps for her kids. One tracks her finances. One manages her family's schedule. And one just built a safe TV app that's now running on her actual television.

She calls it "Before Claw and After Claw." And after seeing what she built, I get it.

This isn't another article about using ChatGPT to generate a worksheet. Jesse built a system. Always-on agents with specific jobs, specific permissions, and specific personalities. All without writing a single line of code herself. Here's exactly how she did it and how you can do the same thing.

Key takeaways:

• Jesse runs 5 specialized OpenClaw agents, each on a separate Mac Mini with strict permission boundaries
• Her homeschool agent Sylvie turns photos of curriculum books into structured lesson plans and custom watercolor illustrations
• Her coding agent Cole built a safe kids TV app and deployed it to a physical TV in 4 days
• She manages physical toy inventory through photos, connecting digital plans to real-world materials
• Total setup cost: roughly $3,000 in hardware, under $50/month in ongoing AI costs

The Problem With "AI for Homeschooling" Right Now

Most articles about AI and homeschooling tell you to paste a prompt into ChatGPT and get a lesson plan. That works for a one-off question. But if you're homeschooling multiple kids at different levels, you need something that knows your curriculum, remembers where each kid left off, and can pull from your actual materials.

A chatbot doesn't do that. It forgets everything the second you close the tab.

Jesse tried the chatbot approach first. She even tried building a "second brain" in Obsidian, logging every lesson and tracking each kid's progress manually. But as a mom of four, she didn't have time for all that data entry. Her second brain was basically brain dead before it even started.

So she went a different direction. Instead of a chatbot she'd use once and forget, she built always-on AI agents that live on her desk and run 24/7.

Approach	Best For	Main Limitation	Ongoing Effort
ChatGPT / Gemini (one-off prompts)	Quick worksheet generation, single questions	No memory between sessions, no file access	Re-explain context every time
AI Homeschool Apps (Kuraplan, Pirch)	Structured curriculum with templates	Locked to their format, can't customize deeply	Monthly subscription, limited flexibility
OpenClaw Agents (Jesse's approach)	Fully personalized, multi-kid, always-on	Hardware cost upfront, learning curve	Under $50/month API costs, agents run themselves

Meet Sylvie: The AI Homeschool Agent

Jesse's first agent handles the biggest time sink in her life: homeschool planning. She named it Sylvie. And the way she set it up is brilliantly simple.

She took pictures of entire curriculum books. Not scanning them page by page. Just snapping photos with her phone and sending them to Sylvie through OpenClaw.

Sylvie didn't just do basic text recognition. The agent understood the structure of the books. The concepts. The lesson progressions. The vocabulary. It turned a pile of physical books into a searchable knowledge base that Jesse can actually query and use.

From there she started asking Sylvie to generate lesson plans from those books. The agent extracts key objectives, vocabulary lists, and materials needed for activities. Dense chapters become one-page teaching guides she can glance at while wrangling four kids.

But here's where it gets really cool. For a lesson on animal survival, Jesse gave Sylvie a photo of the concepts from the book and said: "I want watercolor illustrations suitable for kids that can print on 8.5 by 11 of each of these concepts." Using a Gemini model, Sylvie generated gorgeous watercolor illustrations for things like camouflage, defense, and finding food. Beautiful stuff her kids could actually learn from.

And this detail matters: Jesse wrote a "soul file" for Sylvie. It's basically a personality profile. She made Sylvie a creative, bubbly teacher who is passionate about making learning pop for kids. So every piece of content the agent generates carries that energy. The persona shaped the output.

From Curriculum Books to Custom Kids TV App (In 4 Days)

Jesse's second agent is Cole, the coding agent. And what Cole built is honestly the part of this story that blew my mind.

Like a lot of parents, Jesse was frustrated with YouTube. You know what I'm talking about. Those weird AI-generated videos with strange animations and bizarre storylines that somehow end up in your kid's recommended feed. She wanted a safe, curated viewing experience but had no idea how to build one.

So over four days, mostly working from her phone in small pockets of time between parenting duties, she directed Cole to build an app called Mira.

The concept: an app that pulls from curated YouTube channels to create themed streams of quality content. The interface for kids is dead simple. A go button. Forward. Back. Pause. No comments. No suggested videos. No rabbit holes. Just good content in a locked-down environment.

When she told Cole she wanted it on their actual TV, the agent said it wasn't possible. And this is where Jesse's management instincts kicked in. She literally told Cole: "Try harder. That's not an answer we need right now. We've got real work to do. Save these kids' souls, Cole."

And Cole found a solution using a Google TV Streamer. Guided her through packaging the app and deploying it to a physical device. Now her family TV has a dedicated Mira app that the kids use with a simple remote.

A mom with no coding background built a custom TV app and deployed it to her television. In four days. From her phone. That's not a hypothetical. That happened.

Bridging Digital and Physical: The Toy Inventory System

This workflow is honestly my favorite because it solves a problem every parent has.

Jesse owns tons of educational toys, books, Montessori materials, science kits. They'd sit in cupboards, forgotten. She'd rediscover the perfect dinosaur book long after her son's dinosaur phase had passed. And Sylvie couldn't help with lesson planning around physical materials because the agent didn't know what was physically in the house.

The solution was simple. She took photos of everything. Every educational toy, every book, every set of materials. Sent the photos to Sylvie and said: "I want to make an inventory of my learning supplies."

Sylvie did the rest. Just from images, the agent created a structured inventory in Obsidian. It identified each item, its type, the appropriate age range, and a description. A photo of a wooden board became a catalog entry: "Montessori Language Materials, wooden alphabet tracing board, age range 3-5."

Because the inventory and lesson plans live in the same system, Sylvie can now connect them. When Jesse plans a lesson, Sylvie says: "Hey, you own that tracing board. Maybe pull it out of the cupboard for today's lesson." The agent reaches into her physical world and makes it part of the daily plan.

And one more thing. Jesse gave Sylvie access to her printer. Need a worksheet from a book? Take a photo with her phone. Say "Sylvie, print this." Thirty seconds later the physical worksheet is ready. That tiny friction reduction is massive when you're a parent who literally doesn't have free hands most of the day.

The Security Setup: How to Protect Your Family's Data

Here's what separates Jesse's approach from just throwing everything at one AI chatbot. Each of her five agents runs on a separate Mac Mini. That's not for show. It's a security architecture.

Agent	Job	Permissions	Communication
Sylvie	Homeschool planning	Obsidian vault, printer, image generation	Family chat
Cole	Coding / app building	Code repos, deployment tools	Family chat
Finn	Finance	Read-only bank statements	Private Slack channel only
Claire	Scheduling	iMessage, calendar	Family chat
Agent 5	General tasks	Varies by task	Family chat

Notice what Finn can and can't do. Read-only access to bank statements. Can't send messages outside a private Slack channel. Can't touch the internet. Jesse's scheduling agent Claire can use iMessage but has zero access to financial data. Each agent gets exactly the permissions it needs and nothing more.

This is called the principle of least privilege. It's the same approach enterprise security teams use. And a homeschooling mom figured it out on her own because it just makes sense. You wouldn't give your babysitter your bank login. Don't give your AI agent access to everything either.

When AI Agents Are NOT the Right Choice for Your Family

I'm going to be honest here because I think it matters.

If you homeschool one kid and your biggest pain point is "I need a worksheet on fractions," you don't need a full agent setup. ChatGPT or Gemini will handle that in 30 seconds for free. An always-on agent is overkill for simple prompts.

If your family isn't comfortable with technology running in the background, this might create more stress than it removes. Jesse is tech-curious by nature. She was already using Obsidian. The transition to agents was a natural step for her.

And if your budget is tight, $3,000 in Mac Minis is real money. You can start with one machine running multiple agents, but the security benefits of separate hardware go away. Be honest about what you actually need before investing.

The sweet spot: families homeschooling multiple kids, with lots of physical materials, who want a system that remembers context and runs without constant input. That's where agents shine.

How to Start Your Own Homeschool Agent (Step by Step)

You don't need five Mac Minis to get started. Here's the minimum path:

1. Get one Mac Mini (M-series, any model). Even the base M2 with 8GB RAM works for a single agent.
2. Install OpenClaw. It's free, open source. The install takes about 10 minutes.
3. Write a soul file. This is your agent's personality and instructions. Jesse made Sylvie a "creative, bubbly teacher." You define what fits your family.
4. Connect your tools. Start with Obsidian or a simple folder of documents. Add printer access later. Add image generation later. Layer it up as you get comfortable.
5. Send photos of your curriculum. Just like Jesse did. Snap pictures with your phone, send them to your agent, and ask it to build lesson plans from the content.

The first week will feel clunky. You'll be figuring out how to talk to your agent, what it can and can't do, where it needs more context. That's normal. Jesse described the same learning curve. But by week two, you'll wonder how you ever planned lessons without it.

And if you want to learn alongside other people figuring this out, that's exactly what Shipping Skool is for. We've got parents, small business owners, and first-time builders all sharing what works and what doesn't. No coding background required.

Why the Best AI Users Aren't Programmers

Jesse's story proves something I've been saying for months. The most creative AI agent use cases aren't coming from Silicon Valley engineers. They're coming from parents, small business owners, and people with real friction in their daily lives.

Programmers think like programmers. They optimize for efficiency and architecture. Which is great. But Jesse thought like a mom. She looked at her life, identified the friction points, and asked her agents to fix them. Photographing curriculum books. Building a safe TV app. Creating a toy inventory from snapshots. None of those ideas would come from a developer because a developer wouldn't have those problems.

I think about my own journey with this. I worked minimum wage restaurant jobs for years. I didn't know what a bash command was when I started building with AI. And now I'm running 13 agents on a Mac Mini handling my entire content operation. The tools are there. They're free. And you absolutely don't need a computer science degree to use them.

If a homeschooling mom of four can build a custom TV app and deploy it to her television in four days, what's stopping you?

FAQ: AI Agents for Homeschooling

Can you use AI agents for homeschooling without coding experience?

Yes. Jesse Genet had never opened a terminal before starting with OpenClaw. She built 5 specialized agents by treating them like new employees and giving clear instructions through natural language. You talk to the agent like you'd talk to an assistant. No programming required.

How much does it cost to run AI agents for homeschool?

Hardware cost is roughly $500-700 per Mac Mini. OpenClaw is free and open source. API costs for AI models (Claude, Gemini) run approximately $20-50 per month depending on how much you use them. You can start with a single Mac Mini for under $600 total upfront and under $30/month ongoing.

Is it safe to give AI agents access to family data?

Safety depends entirely on your setup. Jesse runs each agent on a separate machine with strict permission boundaries. Her finance agent has read-only bank access and communicates only through a private Slack channel. Her scheduling agent can use iMessage but cannot access financial data. Follow the principle of least privilege: give each agent only what it needs for its specific job.

What AI tools work best for homeschool lesson planning?

For one-off prompts, ChatGPT and Gemini work fine. For continuous, personalized planning across multiple kids, AI agents on OpenClaw outperform because they maintain context about your curriculum, track each kid's progress, and can integrate with your actual tools like Obsidian, printers, and file systems. They run 24/7 without losing memory.

Can AI agents create custom educational apps?

Yes. Jesse's coding agent Cole built a custom kids TV app called Mira in 4 days, then deployed it to a Google TV Streamer. The app pulls from curated YouTube channels with zero comments, zero suggested videos, and zero rabbit holes. Jesse directed the entire build from her phone with no coding knowledge.

Ready to start building? Join Shipping Skool where over 100 builders (many with zero coding experience) are figuring out AI agents together. Parents, entrepreneurs, and first-time builders all welcome.