Every Level of a Claude Second Brain Explained
Nate Herk | AI Automation ·2026-06-17 ·1 min read
Summary written by us from the video's transcript. The video, and everything in it, is Nate Herk | AI Automation's work.
Learn the five progressive levels of building an AI second brain, from simple file routing to autonomous semantic search and relationship mapping.
Takeaways
- *Start with a clear *claw.md* router to give the AI exact locations for each type of data.
- *Use a markdown wiki (Level 2) to group related content and enable auto‑memory updates without needing a visual tool like Obsidian.
- *Semantic vector search (Level 3) provides meaning‑based retrieval but works best for narrow queries, not full‑context summaries.
- *Relationship mapping creates clusters that reveal how topics connect, moving beyond simple backlinks.
- *Adopt the lowest level that addresses your pain; higher levels add complexity and may not be necessary.
Purpose and Mindset
The second brain stores notes, recordings, and threads so an AI can ingest and retrieve them later. Effective routing ensures both you and the agent can locate information quickly. Start by reverse‑engineering how you’ll query the data—design the structure with the end use in mind.
Level 1: Basic File Routing
A single *claw.md* (or *agents.md* for Codex) acts as a system prompt and router, directing queries to folders like *context*, *projects*, and *decisions*. This level relies on exact‑word searches; if the AI can’t locate a file, it will ask for clarification.
Level 2: Wiki Organization
Add a markdown wiki that groups related files (e.g., YouTube transcripts, meeting notes) under concepts, comparisons, and techniques. The *claw.md* now routes to additional sections such as *references* and *memory.md*, which can be auto‑updated by Claude Code. The visual layer (Obsidian) is optional; the key is that the AI can follow linked pages.
Level 3: Semantic Search
Introduce vector embeddings for chunks of text, enabling meaning‑based retrieval instead of keyword matching. A *vector index* folder stores these embeddings; searches return semantically similar passages. This works well for pinpointing specific rules or excerpts but may miss broader context that a full markdown file provides.
Level 4: Relationship Mapping
Beyond semantic similarity, map explicit relationships between entities—e.g., linking topics A → B → C—to trace chains of information. This creates clusters and graphs that show how concepts interrelate, moving from isolated files to an interconnected knowledge graph.
Level 5: Autonomous Operation
At the highest tier, the second brain operates with minimal prompting, automatically updating memory, routing, and retrieval across all levels. The author notes they haven’t adopted this level yet, emphasizing that you should adopt only the simplest level that solves your current pain points.