Descriptions:
Y Combinator president Garry Tan delivers a keynote at an AI Engineer conference built around a provocative philosophical claim: AGI is not arriving as a singular event announced from a data center—it is already diffusing into personal infrastructure as persistent agents running on individual context. Opening with the story of 17th-century philosopher Baruch Spinoza, excommunicated at 23 yet refusing a bribe to stop his work, Tan uses the concept of conatus—the drive in every living thing to keep going and increase its power to act—as a frame for why personal AI ownership matters for founders and builders.
Tan’s central construct is “personal AGI”: not a $20/month subscription chatbot that resets when you close the tab, but an owned intelligence stack of markdown skill files, scheduled agents, and persistent memory that accumulates context over time and executes work autonomously. He describes YC’s own adoption: non-engineers on the finance and events teams are now building agent pipelines, including a finance staffer who consolidated roughly a hundred Excel workbooks into a single app using an internal agent without writing traditional code. The most concrete demonstration is the talk itself—five days before delivery, Tan’s agent read three Spinoza biographies totaling 1,500 pages, synthesized them into a dated chronology, flagged disagreements between authors, and ranked the ten most tellable moments with delivery notes.
A key technical distinction runs through the talk: latent-space computation (taste, judgment, interpreting vague requests) belongs in the model, while deterministic computation (arithmetic, SQL queries, scheduling 6,000 conference attendees) must be handled by code called from markdown files. Tan argues that nearly every agent failure traces back to conflating these two modes.
📺 Source: Y Combinator · Published August 06, 2026
🏷️ Format: Keynote Launch
