Interview
The AI Company Simulating the Entire Economy | Simile Co-founder & CEO, Joon Sung Park
- Joon-Sung Park, founder and CEO of Simile, predicts that within 2–3 years, the company will run single simulation sessions valued at $100 million.
- Simile operates as a "simulation market" aiming to predict future human behavior by creating a foundation model of human behavior capable of simulating individuals, subpopulations, ecosystems, and markets.
- The company's origin traces to a 2023 "Smallville" simulation featuring 25 AI agents in a town setting that self-organized, held parties, and remembered interactions, demonstrating early agentic workflows with memory, planning, and reflection.
- Simile's agents utilize a "reflection" mechanism—comparable to "shower thoughts"—to summarize high-volume memory data and form higher-level personal traits and worldviews, moving beyond simple recall to simulate personality.
- Unlike frontier AI companies that build super-rational models for coding and math, Simile aims to model human irrationality, biases, and subjective preferences to create "representation at scale."
- The company distinguishes its value proposition from pure prediction (e.g., forecasting sales) by focusing on causal mechanisms and counterfactuals, helping clients determine how to shape the future rather than just anticipating it.
- To achieve this, Simile collects behavioral data through randomized control trials and A/B testing, prioritizing "what people do" over "what people say" to avoid the gaps inherent in web-text training data.
- Simile achieved product-market fit when Fortune 500 leaders, including CVS, reacted immediately to demos, closing enterprise deals within three months despite traditional long sales cycles in the sector.
- In late 2024, Simile validated its models by predicting human behaviors and attitudes with 85% accuracy compared to self-replication, sparking the current "synthetic panel" market.
- Synthetic panels are projected to exceed the current human panel market size within three years, capable of answering 95% of hypothetical market questions that are currently too expensive or impossible to test via traditional methods.
- Simile recently raised a $200 million Series B round (bringing total funding to $300 million in six months), led by Green Oaks and existing insider Index Ventures to accelerate compute and data acquisition.
- The company's co-founders include Joon-Sung Park, Michael Bernstein (ImageNet co-author), and Percy Liang (coiner of "foundation model"), alongside co-founder Lainey Allen, creating a balance of research rigor and commercial execution.
- Park emphasizes a "common denominator" hiring philosophy, seeking candidates who demonstrate extreme ownership and the ability to reinvent themselves across different stages of their careers.
- Recruitment targets researchers who possess two contradictory "superpowers," such as high data rigor paired with creative artistry, or deep short-term paranoia balanced with long-term religious optimism.
- Simile expects the cost of high-end simulations to rise significantly, with single sessions potentially costing $10–$20 million in compute but commanding $100 million in value for enterprise clients.
- Park identifies the primary industry challenge as bridging the gap between research proof-of-concepts and productionizable technology that can be reliably deployed for decision-making.
- Park envisions a future "10 years out" where every individual has a replicable "digital twin" in a simulated world, fundamentally altering governance, corporate structures, and market dynamics.
- Simile views itself as building the "GPU of intelligence units," where the aggregate value lies in simulating the collective intelligence and diversity of society rather than creating a single super-intelligent "CPU."
- Park warns that the AI sector risks becoming overheated with labs lacking a clear vision for global impact, while highlighting robotics, the inference layer, and specialized chip hardware as under-invested or high-value areas.
- The company maintains that an defensible data strategy—specifically the ability to collect hard-to-acquire behavioral data—is the single most critical factor for AI company viability.
- Park cites a transformative moment in his career when Stanford Professor Mary Woetzel provided unsolicited mentorship and access to a research lab, noting that "never forgetting the first believer" is crucial for founders.