By playing, you agree to uListen's Terms and Privacy Policy and YouTube's Terms of Service.

Claude Code + Obsidian: My family’s AI brain

Scott Farrar is a math educator and parent who got tired of AI chats that drift and forget.

Claire VohostScott Farrarguest

Nov 11, 202634mWatch on YouTube

Summaries, chapters, transcripts, and FAQs on this page are made by uListen, not YouTube.

Great conversations run long. Their best ideas don't have to.

uListen is free on iPhone, iPad, Android, Chrome, and Claude.

CHAPTERS

Jump to publisher timestamps

Chapter summaries

Making AI memory transparent: grounding answers in “my truth”
Claire Vo and Scott Farrar open by contrasting opaque AI memory in typical chat experiences with a more transparent approach. Scott explains his goal: a system where he can see and control exactly what the AI knows about his household and decisions.
- •Claire frames the common uncertainty about what tools like ChatGPT/Claude “really know” and remember
- •Scott describes wanting confidence that AI is grounded in his own facts and context
- •He introduces the idea of an AI system that reads/writes Markdown as an explicit memory store
- •They set up the episode’s theme: AI for household management, not just work or coding
From chat to coding agent: using Claude Code + files for ongoing projects
Scott explains the core shift: instead of relying on a single chat thread, he uses a coding agent that can reference files on his computer. This creates continuity across sessions and turns household questions into trackable projects with checklists and updates.
- •Scott describes the friction of dumping context into a chat and worrying about drift
- •He switches to Claude Code so it can read/write Markdown files as persistent memory
- •A household question (like car insurance) becomes a living project file
- •He emphasizes increased confidence and forward motion when the AI produces concrete checklists
Demo workflow: car insurance comparison, next steps, and better agent emails
They walk through how Scott uses his file-based system to manage car insurance quotes and coverage gaps. Scott contrasts this with his previous approach, where complexity often led to procrastination or over-reliance on an agent.
- •Scott starts with a broad question about updating insurance and gathers quotes into the project file
- •The system can identify blockers and propose specific next actions (like emailing an agent contact)
- •Scott says he previously might have avoided reading documents and leaned on trust by default
- •With AI reading the documents plus home context, he asks more detailed questions and communicates more precisely
Building the “household brain” in Obsidian: appliances, serial numbers, and photos
Scott shows how his Obsidian vault provides structured and searchable context for Claude Code. He explains gathering household data—like appliance serial numbers, warranties, and device info—so the AI can answer questions quickly and accurately later.
- •Scott points to the Markdown project file as the “memory” Claude references and updates
- •He uses Obsidian’s graph view to visualize the broader set of household context
- •He stores appliance inventories, serial numbers, and even photos of labels for reference
- •He describes asking Claude what info to gather, then collecting it (including taking pictures of appliances)
Real-life use case: the broken ice maker and warranty-aware troubleshooting
Claire and Scott bond over a shared household problem: broken refrigerator ice makers. Scott explains how the system retrieves relevant appliance context (like warranty dates) and helps him plan next steps without digging through folders or paperwork.
- •Claire pauses to share her own broken ice maker situation and household “adulting” pain
- •Scott describes starting a fresh Claude Code session with a loose question like “What’s going on with the fridge?”
- •The AI consults the stored appliance information and can surface warranty details previously captured
- •Scott compares the system to his dad’s pre-AI file-cabinet approach to receipts and warranties
Type A vs. Type B organization: letting the agent impose structure on messy inputs
Claire proposes that personal organization style shapes AI system design, from highly structured to totally unstructured. Scott argues his approach is a hybrid: he can dump information in, then rely on the coding agent to organize and maintain consistency—supported by automation.
- •Claire contrasts “Notion boy” style organization with her own more unstructured approach
- •Scott describes a spectrum: unstructured capture with agent-driven structuring
- •He mentions a linting script to prevent the system from becoming overly verbose or messy
- •He highlights the power of agents to organize longstanding files on a computer that people never had time to sort
The shared household to-do list: capturing everything, prioritizing, and estimating costs
Scott explains a key artifact he revisits manually: a massive household to-do list built from a brainstorming session with his wife. The list ranges from small fixes to large issues, and the AI helps add structure like priority and cost estimates.
- •Scott and his wife brainstorm every to-do item—big and small—to get it out of their heads
- •Examples include bathroom fixes and evaluating a potentially expensive dry rot issue
- •He asks the system to compile items with priority levels and cost estimates
- •Claire highlights the multiplayer nature of home task management and asks how collaboration works
Multiplayer at home: Dropbox vs. GitHub for syncing the household brain
Scott explains the practical constraints of sharing an AI-backed knowledge base with a spouse. He describes experimenting with Dropbox and considering GitHub, then settling on a hybrid approach given real-world usage patterns and the need for some scripts.
- •Scott’s initial constraint: both spouses should be able to use their own AI with the same knowledge base
- •He tries Dropbox syncing but notes the possibility of sync conflicts if editing simultaneously
- •Claude suggests GitHub as an alternative, but Scott notes the trade-offs for a non-technical collaborator
- •They land on a mixed setup: Dropbox for most content, GitHub for scripts like linting and wall-display exports
Household wall display with DAKboard: surfacing priorities, trips, and “wins”
Scott describes turning the household database into something visible for the whole family via a wall display. He uses DAKboard with a custom webpage, fed by scripts that extract top to-dos, upcoming trips, suggestions, and recent accomplishments.
- •Scott notes Obsidian’s graph view is useful for him but not readable for the household
- •He uses DAKboard and embeds a custom webpage to show actionable information
- •Claude helps build a script to scrape to-do items, trips, and other key info into the display
- •He adds a “recent wins” section to recognize completed tasks and reinforce progress
Does it change behavior? Checklists, reminders, and fewer “pending fires”
Scott reflects on how the system alters his daily execution by converting vague obligations into recorded, revisitable steps. He describes being reminded of overdue items and feeling like he’s moved from endless backlog to waiting on others after sending the necessary emails.
- •Scott says checklists prevent tasks from becoming fleeting “should” thoughts
- •The system can remind him about delays and outstanding items during related chats
- •He reports significant progress across projects and a new feeling of having fewer active fires
- •He characterizes the new bottleneck as external—after he’s done his part and is awaiting responses
Household tasks to hand to AI: repairs, planning, IT, trips, and readiness projects
Scott lists the categories he includes in his household system, spanning practical maintenance to longer-term planning. He gives an example of an “earthquake readiness” question turning into a project with captured research and a checklist-driven workflow.
- •Scott names categories: repairs, insurance-related work, planning items, IT/device management, and trip planning
- •He mentions long-term or complex tasks like consolidating retirement accounts as an area where help is useful
- •He demonstrates starting with a loose prompt about earthquake planning and building a new project from it
- •He explains his structure: ideas can be promoted into projects, completed projects get closed and added to wins/display
AI, kids, and student agency: keeping learners doing the cognitive work
Claire pivots to Scott’s perspective as a math teacher on AI in education. Scott argues for maintaining student agency—using AI to support thinking and interpretation rather than replacing productive struggle with instant answers.
- •Scott emphasizes student agency and the importance of students doing the cognitive effort
- •He contrasts approaches where an agent does tasks for you versus a model where the human remains active in decisions
- •He warns against AI that short-circuits learning by being too helpful or delivering answers too quickly
- •He favors tools that interpret student thinking and encourage elaboration (e.g., “Tell me more about that”)
A flipped tutor model: having the chatbot learn from the student
Scott gives an example of reversing the typical tutoring dynamic. Instead of teaching directly, the chatbot asks the student to explain, then uses follow-up questions to draw out reasoning—aiming to make the interaction supportive without becoming annoying.
- •Scott describes a chatbot that asks the student to teach it (e.g., how to solve a linear equation)
- •The chatbot follows up with “why” questions to deepen reasoning
- •He notes students may share ideas more comfortably with AI than with humans
- •Claire adds that AI can be an infinitely patient private tutor that supports learning through frustration
When AI gets out of sync: restart, ask “why,” and enforce structure
They discuss what to do when the AI behaves unexpectedly or fails to follow the intended process. Scott describes restarting sessions and interrogating the AI’s reasoning, especially because it’s writing into shared memory files that must stay clean and usable.
- •Scott often starts a new session if things feel out of sync
- •He asks the AI to explain its reasoning to ensure it didn’t write incorrect info into the memory files
- •He gives an example where the to-do file became overly verbose with updates, violating his intended structure
- •Claire contrasts her own “why” style and wraps with takeaways on household context and learning without answer-shortcuts
Where to find Scott and closing remarks
Scott shares where people can connect with him and what he likes to discuss. Claire closes the episode with standard show notes and ways to follow the podcast.
- •Scott shares his website and mentions LinkedIn as a place to find him
- •He says he’s eager to talk about math education
- •Claire thanks him and closes with ways to like, subscribe, and find the show on podcast platforms

Publisher timestamps

Get more out of YouTube videos.

Read along on your phone, search it beside the video, or ask your favorite AI app what was said.