Claude will handle it. Local. In one folder. (No Obsidian Needed)
The creator argues that using Claude Code with a simple local folder structure is superior to the popular Obsidian + Claude setup promoted by Andrej Karpathy's viral post. He demonstrates his own personal knowledge assistant system built entirely in VS Code with local markdown files and AI agents. He contends that Obsidian adds unnecessary complexity and file-locking constraints without meaningful benefits.
Summary
The video is a response to the viral trend of combining Obsidian with Claude for personal knowledge management (PKM), largely sparked by Andrej Karpathy's post that received 17 million views. The creator argues this trend overcomplicated a simple problem and that Obsidian is unnecessary when Claude Code and VS Code with a local folder structure can accomplish the same goals with more flexibility.
The creator begins by breaking down Karpathy's setup: a 'raw' directory for ingesting articles and papers, which gets compiled into a wiki with backlinks, and Obsidian used as the visual interface. He argues that Obsidian's role is merely as a reader/viewer, and since the LLM writes and maintains all the wiki data anyway, the elaborate Obsidian interface is overkill. He suggests NotebookLM as a better alternative for the specific use case of loading hundreds of articles and querying them.
A central concern raised is scalability and consistency: loading raw articles without structured indexing means the AI starts from scratch each time, likely producing different answers to the same question. The creator argues that a proper knowledge base should contain curated conclusions and signal, not raw noisy article dumps.
The creator then demonstrates his own setup, which uses VS Code to access a local folder structure with clearly defined 'inboxes' — a team inbox where he drops items for AI processing and an owner's inbox where the AI deposits outputs for his review. He emphasizes that VS Code, unlike Obsidian, does not alter or lock markdown files to a proprietary format.
He reveals a more advanced personal setup: a custom web interface he vibe-coded himself that includes daily journaling, PDF document management, a database explorer, backlinking, mood tracking, and agent management — capabilities he argues Obsidian could never provide without complex third-party plugins. His business setup goes further, featuring a multi-agent AI team with specialized roles (Larry the orchestrator, Jack the community manager, etc.), connected to external databases including their membership platform with thousands of pieces of content.
The creator concludes by addressing comments from Andrej's post and emphasizes that the local folder is the actual knowledge base — not Obsidian, not Claude Code. Obsidian is just one possible viewer, and an inferior one compared to custom-built interfaces. He encourages viewers to watch a separate video where he builds the entire system from scratch using only a local folder and Claude Code in a terminal.
Key Insights
- The creator argues that Obsidian's role in Karpathy's setup is purely as a viewer, and since the LLM writes and maintains all wiki data directly, having Obsidian as an interface is 'already overkill' — the interface adds no value when you rarely touch the data manually.
- The creator claims that loading raw articles into a knowledge base without structured indexing is fundamentally flawed at scale, because the AI starts from scratch each time and will likely give different answers to the same question — arguing that only curated conclusions and signal, not raw noisy content, should populate a real knowledge base.
- The creator distinguishes VS Code from Obsidian by pointing out that VS Code gives access to local files without altering them, while Obsidian modifies files to fit its own database format, effectively locking content into the Obsidian ecosystem.
- The creator demonstrates a custom vibe-coded personal interface that includes journaling, PDF management, database exploration, mood tracking, and Stoic philosophy feedback — arguing this is what a 'real personal knowledge base' looks like versus a simple wiki, and that Obsidian could never provide this without complex plugin dependencies.
- The creator reveals that his business operates with only two human team members (himself and co-founder Paco), with AI agents completely replacing other human roles — including an orchestrator agent named Larry who delegates tasks to specialized agents and can autonomously 'hire' new agents when a task falls outside existing team capabilities.
Topics
Transcript
[0:00] Using AI for personal knowledge management is so much simpler. Then I see other people are setting it up and in this video, we'll show you proof. So you might have seen all these videos about , use Claude with obsidian, and I get it why they're coming up because obsidian has a following base and a lot of traction. And using this with AI now seems to be the right thing to do. And many of these videos are based on a post on X from Andrej Karpathy, also the inventor of the term vibe coding. Um. [0:30] Who came up with this LLM knowledge basis, and if you watched my previous videos long before this post, I…
Full transcript available for MurmurCast members
Sign Up to AccessMore from ICOR with Tom | AI Productivity
Obsidian is the tool I teach in. Keep the one you have.
The creator addresses criticism about switching to Obsidian after previously promoting other tools, clarifying that his tool-agnostic ICOR methodology remains unchanged and that Obsidian is simply an interface layer for his folder-based system. He emphasizes that the underlying folder structure, AI automation, and productivity principles are identical regardless of which tool visualizes the content.
Claude sucks. And here is why.
This video explores why Claude and other AI models produce inconsistent results, examining how model selection (Haiku vs Sonnet vs Opus vs Claude 3.5 Fable), effort levels, prompt clarity, and system context dramatically affect output quality. The speaker demonstrates these differences through a Venn diagram creation task and advocates for using organized folder systems with documented SOPs and code-based solutions to ensure consistent AI performance.
Claude can now prompt its own SESSIONS. I built the boss.
Claude now has the ability to interact with multiple sessions, allowing a streamlined orchestration of tasks through an agent called Larry. This integration can simplify task management and enhance productivity by enabling a single main session to oversee various sub-sessions or tasks.
I switched off Claude's memory. My folder does it better.
Tom explains why he disabled Claude's native memory feature and instead built a local folder-based system (myPKM) to maintain persistent context with AI agents. He shares an audit revealing systemic issues like contradictory documentation, forgotten rules, and parallel session conflicts, then outlines how to systematically fix these problems using the ICOR productivity methodology.
I do not download invoices anymore. My folder and Claude do it.
The speaker demonstrates how he uses AI (Claude) integrated with his local folder system to automate invoice collection for his accountant, downloading 36 invoices in 5 minutes with minimal manual intervention. He explains his AI team methodology, work streams, and how persistent local knowledge bases enable AI to handle recurring tasks independently without constant re-explanation.