Summary
Jun Park, co-founder and CEO of Simile, sits down with Harry Stebbings on 20VC to discuss how the best AI companies differentiate themselves through proprietary data strategies and simulation capabilities. Park traces the origins of Simile back to a 2023 Stanford research project that populated a virtual town with 25 AI-driven non-playable characters — one of the earliest demonstrations of autonomous agents with explicit memory, planning, and reflection capabilities, built on top of GPT-3.5 before ChatGPT existed.
In the conversation, Park explains how those early agent architectures — where agent memory was initially stored in simple markdown files — evolved into the foundation for Simile’s commercial simulation platform. The company is co-founded alongside Stanford researchers Michael Bernstein, co-author of the image generation paper widely credited with sparking the modern AI revolution, and Percy Liang, who coined the term “foundation model.” Their thesis: as models get better at representing human behavior, simulation fidelity improves directly, creating tight alignment between research progress and product value.
Park also discusses Simile’s enterprise go-to-market, including work with CVS, and reflects on how corporate adoption of simulation technology came far faster than his original late-2026 target. The interview offers rare insight into how research-lab origins, defensible data acquisition, and product-research alignment can form a durable moat in an increasingly competitive AI landscape.
📺 Source: 20VC with Harry Stebbings · Published August 01, 2026
🏷️ Format: Interview
