How I AI5 OpenClaw agents run my home, finances, and code | Jesse Genet
CHAPTERS
- 0:00 – 5:14
From Obsidian “second brain” to OpenClaw: why Jesse jumped in
Jesse explains how she discovered OpenClaw through Obsidian power users and why it clicked for her life as a homeschooling mom of four. The core motivation is reducing the manual work of organizing information so she can actually use her “second brain,” not just aspire to it.
- •Found OpenClaw via an Obsidian influencer comment about agents using your files
- •Wanted an AI layer to make Obsidian practical under real parenting constraints
- •Goal: run homeschool planning and logging through an agent-driven workflow
- •Framing: “before Claw” vs “after Claw” as a life/ambition unlock
- 5:14 – 5:44
Layering OpenClaw on top of an Obsidian homeschool vault
Jesse walks through her Obsidian vault structure (“Family Learning”) and what she’s been trying to capture about her kids’ education. The key is turning day-to-day life into usable, queryable data without demanding constant manual note-taking.
- •Obsidian vault as a set of markdown files used as a family learning system
- •Tracks more than lessons (e.g., trips and educational moments)
- •Desired structure: date, instructor, children involved, what was taught, next steps, notes
- •OpenClaw provides the missing labor to keep the system updated
- 5:44 – 7:27
Automatic lesson logging from photos to enable adaptive curriculum
She describes a workflow where she takes photos of lessons/activities and has OpenClaw generate structured logs. Those logs then become the substrate for personalized curriculum suggestions based on each child’s progress over time.
- •Capture lessons via quick photos instead of manual writing
- •Agent extracts lesson details into structured entries
- •Progress tracking across multiple children and ages
- •Using the log history to recommend next activities tailored per child
- 7:27 – 10:30
Turning entire books into structured, teachable curriculum assets
Jesse explains how she feeds reference books into the system (often by photographing pages) so the agent can generate lesson plans and supporting materials. They discuss a specific phonics book and how AI can make dense curricula easier to teach and remix.
- •Curriculum sources vs curriculum plans: separating references from generated lesson progressions
- •Photographing books so the agent can extract and transform content
- •Example: phonics book challenges and AI-generated lesson scaffolding
- •Ideas: extract stories, generate printouts, even create physical learning aids (e.g., 3D printed letter forms)
- 10:30 – 13:06
Generating complete lesson plans and custom illustrations from snippets
Jesse demonstrates how OpenClaw turns a curriculum chapter into a full lesson plan with objectives, vocabulary, materials, and activities. She also shows creating printable watercolor-style illustrations from a simple prompt plus a photographed page of concepts.
- •Auto-generated lesson plan format: objectives, concepts, vocab, materials, activities
- •‘Lo-fi’ input: photo of a book snippet becomes structured outputs
- •Image generation integration (API key + model) to create kid-friendly visuals
- •Emphasis on lowering prep time: teach tomorrow without reading the whole chapter
- 13:06 – 14:40
SOUL.md personalities: designing an agent as a world-class teacher
Jesse introduces SOUL.md files to shape each agent’s persona and behavior, describing Sylvie as a bubbly, creative homeschool teacher. The chapter highlights how persona design changes output quality and makes specialized agents feel more effective.
- •Multiple agents with distinct roles feel more natural and productive
- •Sylvie’s persona optimized for kids’ learning and creativity
- •SOUL.md influences tone, priorities, and how prompts are interpreted
- •Combining simple user prompts with persona-driven “extra” instructional flair
- 14:40 – 17:06
Why five specialized agents—and the reality of collaboration challenges
They unpack Jesse’s five-agent setup and why she prefers deep specialization (homeschool, finance, scheduling/EA, dev, etc.). Jesse is candid that agent-to-agent collaboration is still hard because human communication tools (Slack/Telegram/etc.) aren’t native for agents.
- •Role-based agent design: each agent has a job and a data scope
- •Specialization avoids mixing contexts (e.g., homeschool teacher vs receipts)
- •Moved agents into Slack hoping for better coordination
- •Key limitation: today’s comms channels are built for humans, not agent interoperability
- 17:06 – 18:30
Slack setup is the hard part: building custom bot access
Jesse explains that adding OpenClaw agents into Slack required creating custom Slack apps, which was more difficult than spinning up the agent itself. This chapter focuses on the practical friction and why “agent orchestration” often breaks on integrations.
- •Slack as a collaboration hub sounds ideal but is nontrivial to set up
- •Creating a custom Slack app/bot per agent was a major hurdle
- •Integrations (Slack/iMessage/Signal/Telegram) are ‘hacks’ around human-first tooling
- •Operational overhead grows quickly as you add more agents
- 18:30 – 21:11
Physical partitioning with multiple Mac Minis for security and trust boundaries
Jesse reveals her desk full of Mac Minis and explains why she physically separates agents. The driving concern is data security and preventing accidental leakage across contexts—especially for financial information.
- •Uses separate machines to isolate worlds (finance vs scheduling vs homeschool)
- •Finance agent may access sensitive statements; scheduling agent has outbound messaging
- •Physical partitioning as a ‘lazy but safe’ method vs more complex software isolation
- •Progressive trust model: limit communication channels and access by default
- 21:11 – 26:13
Managing agents like employees: onboarding, permissions, and progressive trust
Both hosts compare agent management to hiring and onboarding an executive assistant. They emphasize not granting blanket access (email, passwords) and instead provisioning limited permissions, separate accounts, and gradual expansion of responsibilities.
- •Treat agents like new hires: staged access, clear responsibilities
- •Avoid impersonation; agents should have their own accounts/addresses
- •Examples: read-only email access, separate agent email, delegated calendar access
- •Mindset shift: operational security and trust-building over time
- 26:13 – 31:52
Cole the dev agent: building a custom YouTube experience to avoid ‘AI slop’
Jesse demos a bespoke app (“Mira”) built with her coding agent to curate safe, high-quality YouTube streams for kids. The product constrains controls to simple playback/skip, maintains history, and runs on a TV device so kids can’t wander off-platform.
- •Motivation: protect kids from misleading/low-quality recommended content
- •Agent-built app creates themed, endless streams without manual playlists
- •Kid UI constraints: play, next/previous, pause; history preserved
- •Deployment to TV via Google TV streamer so the app is effectively locked-in
- 31:52 – 35:54
The ‘after Claw’ time revolution for parents: async work and voice-driven control
They discuss how having agents operate the computer lets parents work in small, fragmented windows without being chained to a laptop. Voice-to-text and asynchronous agent execution become critical when your hands are literally unavailable (e.g., holding a baby).
- •Parents’ schedules are fragmented; agents enable micro-sprints and async progress
- •Voice notes + agents running tasks on your machine reduce screen time
- •“Hands problem” framing: both humans (parents) and agents lack physical hands
- •Emotional impact: reclaim ambition while staying present with kids
- 35:54 – 41:02
Bridging into the physical world: cupboard photos → household learning inventory
Because agents can’t physically organize cupboards, Jesse feeds them visual context by photographing shelves and materials. OpenClaw converts photos into a structured inventory, then links items to lesson plans so she knows what to pull out when.
- •Workaround for no physical body: create visual inputs of physical spaces
- •Photo-to-inventory extraction: category, age range, description from images alone
- •Connect inventory to existing lesson plans for just-in-time material recommendations
- •Extends beyond homeschool: batteries, appliances, books, toys—search your house via AI
- 41:02 – 49:27
Printing from voice: the ‘low-tech’ killer workflow and memory/decision files
Jesse highlights a surprisingly transformative feature: her agent can print to a normal printer from a voice instruction, eliminating multi-step scanning/emailing hassles. She closes with how she manages agent behavior using decision files and SOUL.md updates—mostly by having the agent self-diagnose rather than manually editing.
- •Agent can trigger real-world actions via computer peripherals (printing)
- •Instant workflow: photo a worksheet/page → ‘print this’ → physical paper quickly
- •Memory scaffolding: decisions files to lock choices and avoid re-litigating tasks
- •Behavior tuning via commands like ‘That’s a decision’ and ‘Update your SOUL.md’
