YC CEO Garry Tan shares how to build your AI "second brain," fat skills + thin frameworks (free download pack included)

📄Full Article· Automatically extracted by trafilaturaGemini 翻譯4773 words

Y Combinator CEO Garry Tan shared on X how he uses a "Fat Skills, Fat Code, Thin Framework" architecture to build a compounding personal AI system: 100,000 pages of knowledge base, 100+ composable skills, book mirrors, and the Skillify loop that lets skills create skills. (Context: Anthropic released 10 financial AI agents: reading earnings calls and financial reports, creating Pitchbooks, and month-end closing; Claude does it all for you.) (Background: Getting desperate? OpenAI attacks Anthropic for inflating annual revenue to $8 billion, claiming actual income is less than ChatGPT.) People keep asking me why I’m still coding at 2 AM every day. I have a very heavy job: CEO of Y Combinator, helping thousands of founders build real startups every year, pursuing real revenue and rapid growth. Over the past five months, AI has made me a creator again. By the end of last year, the tools were good enough for me to go back to building things. Not toy projects, but systems that truly compound. I want to show with concrete examples what a personal AI looks like when you stop using it as a chat window and start using it as an operating system. I’ve open-sourced all of this and written it up because I want you to keep up with my speed. This is part of a series: "Fat Skills, Fat Code, Thin Framework" introduces the core architecture; "Parsers" covers intelligent routing tables; "LOC Controversy" discusses how every technologist can multiply themselves by 100x to 1000x; "Bare Models are Dumber" argues that the model is the engine, not the car; and the "Skillify Manifesto" explains why LangChain raising $160 million gave you a set of weight-training equipment but no training plan. Last month I was reading Pema Chödrön’s *When Things Fall Apart*, 162 pages, 22 chapters, discussing Buddhist perspectives on suffering, rootlessness, and letting go. A friend recommended it to me during a difficult time. I asked the AI to create a "Book Mirror." Specifically: The system extracts all 22 chapters of the book and simultaneously executes sub-agents for each chapter: summarizing the author's ideas, then mapping each idea to my real life. Not generic "this applies to leaders" nonsense, but specific mappings. It knows my family history (immigrant parents, dad from Hong Kong and Singapore, mom from Myanmar). It knows my professional background (running YC, building open-source tools, mentoring thousands of founders). It knows what I’m reading, what I’m thinking about at 2 AM, and what I’m exploring with my therapist. The output is a 30,000-word report. Each chapter is presented in two columns: what Pema said, and how it maps to what I am actually experiencing. The chapter on rootlessness links to a specific conversation I had with a founder the week before. The chapter on fear maps to a pattern my therapist identified. The chapter on letting go cites notes I wrote late at night about the creative freedom I found this year. The whole thing took about 40 minutes. A therapist charging $300 an hour couldn't do this in 40 hours because they don't have my professional background, reading history, meeting notes, and founder relationships loaded and cross-queryable. I have now read over 20 books this way: *Amplified* (Dion Lim), Bertrand Russell’s autobiography, *Designing Your Life*, *The Gifted Kids' Playbook*, *Finite and Infinite Games*, *Gift from the Sea*, *Siddhartha* (Hesse), *Steppenwolf* (Hesse), *The Art of Doing Science and Engineering* (Hamming), *The Dream Machine*, *The Book: On the Taboo Against Knowing Who You Are* (Alan Watts), *What Do You Care What Other People Think?* (Feynman), *When Things Fall Apart* (Pema Chödrön), *A Brief History of Everything* (Ken Wilber), and more. Each one becomes more valuable because the "brain" gets richer. The second mirror knows the first exists, and the twentieth knows everything about the previous nineteen. The first book mirror was terrible. The first version had three factual errors about my family: it said my parents were divorced, but they weren't; it said I grew up in Hong Kong, but I was actually born in Canada. So I added a mandatory verification step. Now, every mirror performs a cross-modal evaluation against known facts in the "brain" before outputting. Opus 4.7 1M catches precision errors, GPT-5.5 catches missing context, and DeepSeek

Data Status✓ Full text extractedRead Original (動區 BlockTempo)

🔍Historical Similar Events· Keyword + Asset Matching1 items

2026-05-10

Trust math, not ideology: Strategy CEO defines the only condition for selling BTC

Similarity 120%關鍵字 ceo同分類 zh

💡 Currently matching via keywords + symbols (MVP) · Will be upgraded to embedding semantic search later

Raw Information

ID:a003dcc370

Source:動區 BlockTempo

Published:2026-05-12 07:02:33

Category:zh_news · Export Category zh

Symbols:Unspecified

Community Votes:+0 / −0 · ⭐ 0 Important · 💬 0 Comments