Garry Tan
CEO of Y Combinator; argues personal AI should be a compounding operating system made from fat skills, fat code, thin harness, and a fat data layer.
Garry Tan is the CEO of Y Combinator and the author of the X Article "Meta-Meta-Prompting: The Secret to Making AI Agents Work." In the article, he describes returning to hands-on building because AI agents made it practical to create personal systems that compound across work, reading, meetings, and relationships.1
Tan's core claim is that effective personal AI is not a chat window but an operating system: a thin harness routes work to fat skills, fat code, and a fat data layer. The model is treated as an interchangeable engine, while the durable value lives in the accumulated Self-Improving Knowledge Base, reusable skills, structured context, and automation loops.1
His examples include book mirrors that map chapters to his own life context, meeting prep for a Demis Hassabis fireside chat, entity propagation after meetings, and daily crons for social media, Slack, email, and knowledge-graph updates. These examples connect Personal Agents, Harness Engineering, Skillification, GBrain, and Hermes Agent into a single compounding-system thesis.1
Mentioned in 2 pages
Engineering with agents
People & organisations