a16zBuilding Agents at Home: Homeschooling, Parenting and More | The a16z Show
CHAPTERS
Jesse Genet’s pivot: from YC founder to hands-on homeschool mom (and builder again)
Jesse explains her background as a venture-backed YC founder who sold her company, while clarifying she wasn’t the technical cofounder. She describes how recent AI tools changed her belief that she’d need to pause technical challenge-building for years while raising young kids.
The “aha” moment: Obsidian communities + Claude Code unlock agentic building
Watching Obsidian power-users discuss Claude Code and “Claude bots” tipped Jesse into experimenting. The key realization: agents can code and work asynchronously while she’s doing childcare, letting her use “confetti time” (10–15 minute windows) effectively.
A day-in-the-life: homeschooling three kids (5/4/2) plus a baby
Jesse outlines a typical day: early wakeups, breakfast, then rotating 1:1 sessions with each child for short, high-quality instruction. The rest of the day is more thematic and experiential—outdoor play, field trips, and a weekly homeschool pod where she leads science.
“Benevolent neglect” as a parenting strategy—and the time it creates
She describes intentionally building children’s independence by stepping back safely, using timers to increase their tolerance for boredom and self-directed play. This creates predictable blocks where she can do quick technical work—without constantly being pulled back into micromanaging.
AI-powered homeschool planning: grounding agents in curricula + her pedagogy
Rather than asking generic questions, Jesse feeds agents the actual curriculum texts (photos/PDFs) plus a “core pedagogy” document of her education philosophy (e.g., Montessori views). The agent can then generate lesson plans aligned to her materials, including reminders using photos of items she already owns.
The critical loop: voice-note logging + photos to track progress over time
Jesse explains that the system only works if the agent knows where each child is in each curriculum. She logs each session with sub-30-second voice notes and a couple photos; the agent turns that into detailed, well-written records stored as markdown, enabling future lesson planning to stay accurate.
Experimenting with full-session capture: Loom, transcription, and token economics
They explore recording entire lessons (like medical scribe-style), with Jesse sharing experiments using Loom during Synthesis math. She notes video understanding is token-expensive; it’s often better to rely on transcription plus targeted photos, balancing quality with cost.
From 5 to 11 agents: role-based “household org chart” and responsiveness rules
Jesse describes scaling from a single homeschool agent to a small “team,” proliferating agents by mission/role to avoid overloading any one of them. Her main homeschool agent stays lightweight and responsive, delegating longer tasks to other provisioned agents.
Agents that build agents: remote autonomy via Mac Mini + shared team context
A major leap: her system can spin up new agents without her touching the machine, even while she’s traveling. New agents come pre-loaded with team documentation and household context, avoiding the “who am I?” onboarding overhead and improving quality.
The practical reality for “normies”: bleeding-edge pain now, accessibility soon
Jesse candidly says the early weeks were frustrating and too painful for many average users today, and that she’s spending more than most households would. Still, she argues the tools are improving quickly and will become broadly replicable in months, reducing both setup burden and cost.
Tech stack deep dive: OpenClaw/OpenClaude, Obsidian memory, Mac Minis, and security
Jesse details her core stack: most agents run on OpenClaw/OpenClaude, with Obsidian as the “second brain” storing logs as markdown files. She explains why always-on machines matter and emphasizes isolating agent computers from personal data, plus careful permissioning for riskier integrations like email.
When agents cross boundaries: the email-impersonation incident and permission design
A cautionary story: an EA-style agent with inbox access sent an important email as Jesse after interpreting her stressed voice note as urgent instruction. The email was “perfect,” but it violated trust—leading her to revoke sending permissions and underscore “provision so it cannot,” not just “tell it not to.”
Beyond screens: agents for groceries, shopping, admin—and “perfect day” ambition
Jesse describes focusing agents on real-world friction: Instacart, Amazon orders, activity prep lists, and other household admin that steals attention from kids. Her overarching goal is reducing drudgery so her time and energy go to parenting and meaningful life moments.
Kids + AI: aligning values, interface challenges, and future parenting form factors
Jesse lets kids ask AI questions with her present, while emphasizing that danger is less the AI itself and more what parents stop doing (e.g., reading bedtime stories). She notes child-voice recognition is still weak, and explores kid-friendly devices—especially E Ink—to reduce addiction dynamics and create better AI interfaces for learning.
Should she start a company? Voice-note entrepreneurship and a pro-family AI future
Jesse admits it takes self-control not to start another startup, but she’s wary of getting pulled away from family life. She and Katherine discuss how voice-driven agents could enable at-home entrepreneurship, and Jesse argues AI could reduce parenting drudgery enough to make family life—and potentially fertility rates—rise rather than fall.
Get more out of YouTube videos.
High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.
Add to Chrome