@berryxia: YC CEO Garry's knowledge compound interest effect is like a snowball! The system is open-source and free, with clear logic. Highly recommended! Here's the real reason Garry Tan (YC CEO) was coding until 2 AM! AI has turned him back into a builder...

YC CEO Garry's knowledge compound interest effect is like a snowball! The system is open-source and free, with clear logic. Highly recommended! Here's the real reason Garry Tan (YC CEO) was coding until 2 AM! AI has turned him back into a builder, personally creating a personal AI operating system with compound interest effects—GBrain. He no longer treats AI as a chat window, but as a true "second brain / nervous system." Over the past 5 months, he has used this system to complete deep mirroring of 20+ books, automatic prep for every meeting, automatic entity propagation and updates, and the continuous growth of 100,000 pages of structured knowledge. Core Case Breakdown: 1. Book Mirror (The Book Reads Me): Taking Pema Chödrön's "When Things Fall Apart" as an example, the system extracted all 22 chapters. For each chapter, it simultaneously generated the author's viewpoints + precise mappings tailored to Garry's personal life (immigrant family background, work at YC, conversations with therapists, conversations with founders). The entire process took 40 minutes, outputting a 30,000-word two-column comparison. Each subsequent book automatically calls upon previous mirrors, making the context richer and deeper. 2. Meeting Prep (Automatic Meeting Preparation): Before Demis Hassabis came to chat with YC, within 2 minutes, the system pulled up Demis's complete brain page, his views on the AGI timeline, key points from his latest biography, overlaps/divergences with Garry's viewpoints, plus 3 demo scripts and conversation hooks. 3. Skillify Meta-Skills: Whenever faced with repetitive workflows, he simply says "Skillify it," and the system automatically distills a one-time manual operation into a reusable skill (including triggers, edge cases, and tests), registering it in the resolver. Over 100 skills like Book Mirror, Meeting Ingestion, and Enrich were iterated in this way. The architecture is extremely simple and clear: - Thin Harness: OpenClaw / Hermes Agent, responsible only for routing, with just a few thousand lines of code. - Fat Skills: Each skill is a markdown file, focused on a single task, composable and iterable. - Fat Data: 100,000 pages of structured brain pages (people, companies, meetings, books, articles all linked), with 100+ cron jobs updating automatically every day. - Switchable Models: Opus 4.7 1M for precision, GPT-5.5 for recall, DeepSeek V4-Pro for creativity, automatically selected based on the skill. Garry has open-sourced the entire stack: GBrain + GStack + OpenClaw / Hermes Agent, including the coding framework that has garnered 87k+ stars on GitHub. His core point is: The future belongs to those who build AI into compound interest systems, not just to those who use big tech chat tools. The difference is like a diary versus a true nervous system. For all builders who want to truly integrate AI into their workflows, manage personal knowledge, or build Agent systems, this long-form article + open-source project is packed with干货 (practical content). It's worth reading and forking immediately.