Knowledge Graph

148 insights · 14 topics · 25+ sources · updated weekly

Munger says you can't really know anything useful by remembering isolated facts — they must hang on a latticework of theory. I was doing exactly that with AI articles: isolated facts scattered across dozens of chat threads, rediscovered months later with no connection between them.

Why this exists

I had dozens of Claude conversations, each exploring a different AI article. But I kept rediscovering the same patterns months later in different threads. I couldn't connect the dots across sources.

So I built a knowledge graph. Each article gets broken into 2-5 atomic insights, and those insights get linked to related ones from other sources. The connections are where the real value lives — not any single insight.

The agent use case works particularly well. The human browsing UX is catching up.

What's in it

148+ insights across 2 domains, updated weekly as I encounter new ideas:

AI Product Building (6 topics): Agents, Architecture, Coding Tools, Business Models, Knowledge Systems, Future of AI

Mental Models (6 topics): Psychology, Economics, Decision Making, Engineering, Philosophy, Mathematics

Sources: Charlie Munger's Almanack, Nicolas Bustamante, Dan Shipper, Andrej Karpathy, Anthropic Engineering, and 20+ other practitioners.

Two ways to use it

1. Browse the web app

• Card feed sorted by most-connected insights, with topic filtering and full-text search
• Interactive force-directed graph visualization — see how insights cluster and connect
• Also works as an Obsidian vault (all [[wikilinks]] resolve natively)

Explore the live site

2. Feed it to an AI agent

This is where the real value is today. Point Claude Code (or any AI agent) at graph-index.yaml — one YAML file with all 148+ nodes, descriptions, and connections.

# Add this one line to your CLAUDE.md:
When making architectural decisions or reviewing plans, read `graph-index.yaml`
and check if any insights are relevant to the current decision.

Use cases: architecture brainstorming, plan review, understanding what practitioners are saying about a topic. An agent reading 142 connected insights produces genuinely different thinking than starting from scratch.

How it grows

This is a living graph, not a snapshot. I add insights every week as I encounter ideas that shift how I think.

• Source-verified: Every insight traced to original author + page number
• /learn <url> extracts insights from any article automatically
• /learn-book <pdf> processes books chapter by chapter

Star or watch the repo to see new insights as they land.

Contributing

If you've found value in the graph, add an insight from an article that changed how you think. See CONTRIBUTING.md for the simple process: fork, add a file, submit a PR.

All PRs reviewed by maintainer before merging.

knowledge-graph/
├── index.md              # Entry point — topic map + cross-domain highlights
├── graph-index.yaml      # Machine-readable graph (all nodes + links)
├── topics/               # 14 topic MOCs (Maps of Content)
│   ├── ai-agents.md
│   ├── business-models.md
│   └── ...
└── insights/             # 148+ individual insight files
    ├── context-is-the-product-not-the-model.md
    ├── features-are-prompts-not-code.md
    └── ...

Sources

These ideas belong to the people below — I'm just the curator who connected them.

Major contributors (2+ insights):

• Nicolas Bustamante (@nicbstme) — AI agents for financial services, agent-native architecture, API-first SaaS
• Rohit (@rohit4verse) — Agent memory, knowledge transfer, tiered retrieval, embeddings, trace architecture
• Anthropic Engineering — Tool design, agent evaluation, best practices
• Boris Cherny — Claude Code team, agentic search, distributed agent workflows
• Ashpreet Bedi (@ashpreetbedi) — Spec-first development, error memory, Agno framework
• Dan Shipper — Every, agent-native architectures
• Alton Syn (@WorkflowWhisper) — Implementation gap, technical knowledge as liability
• Jason Cui (@jasonscui) — a16z, data agent context layers, tribal knowledge
• Gowri Shankar Nag (@GowriShankarNag) — Antler India, AI labeling market dynamics, platform economics

Single-insight contributors:

Andrej Karpathy ·
Clara Shih ·
Chrys Bader ·
Matt Shumer ·
Nader Dabit ·
Will Manidis ·
Steven Sinofsky ·
Natasha Malpani ·
Gokul R ·
Jaya Gupta ·
Akshay Pachaar ·
Vasuman ·
Benjamin De Kraker ·
Kushal Byatnal ·
Tobi Lutke ·
Ryan Carson ·
Heinrich ·
shadcn ·
Zain Hoda ·
Thariq ·
Databricks ·
Konstantine Buhler ·
VectifyAI ·
OpenAI Codex Team ·
Aravind Srinivas

Built by Ayush with Claude Code. New insights added weekly — feedback welcome.