In February 2026, competition in the AI agent sector has intensified across the board. Recently, core members of the founding team behind the groundbreaking "Stanford AI Town" officially announced their new venture. The new company, Simile, has completed a $100 million Series A funding round, aiming to build a scalable "AI social simulator" capable of replicating human behavior.
This funding round attracted not only leading investment from top-tier venture capital firms but also endorsements from prominent Silicon Valley figures, including renowned Stanford University AI scholar Fei-Fei Li and OpenAI co-founder Andrej Karpathy.
Simile's technological foundation stems from the influential 2023 paper "Generative Agents." At that time, a virtual town called "Smallville," created by Joon Sung Park, his advisor and current Simile Chief Scientist Percy Liang, and others, demonstrated that just 25 GPT-based AI agents with simple identity settings could autonomously engage in social gossip, information dissemination, and even plan a Valentine's Day party. This research was considered a pioneering achievement in the field of generative agents, proving for the first time that large language models could simulate near-human behavior in a virtual world.
Today, Simile's ambitions extend far beyond pixel-based games. The team has upgraded the original architecture, which supported only 25 agents, into a large-scale simulation platform capable of hosting thousands of agents. The goal is to build a "foundational world model" to predict human behavior at any scale and under any circumstances. Simile's co-founders include Joon Sung Park, Stanford Professor Michael Bernstein, Percy Liang, and Lainie Yallen, with core members possessing strong academic backgrounds from Stanford.
Disclosed information indicates this funding round was led by the well-known venture firm Index Ventures, with participation from Bain Capital Ventures, A*, and Hanabi Capital. Beyond institutional investors, the list of individual investors is notable: Stanford AI Lab co-director Fei-Fei Li, former Tesla AI director and OpenAI co-founder Andrej Karpathy, and Quora founder Adam D'Angelo all participated.
Commenting on Simile's direction, investor Karpathy stated: "This is exploring a very interesting and still underestimated dimension of large language models. The native form of a pre-trained large model is more like a simulation engine trained on vast and diverse human text. Why not leverage this statistical power to attempt simulating a population?"
Simile's core value proposition is providing businesses and organizations with a low-cost "decision flight simulator" for testing. Founder Joon Sung Park explained the venture's premise: "Pilots don't train with real passengers, and surgeons don't practice on live patients. Yet, products, policies, and choices affecting millions in society are often launched directly into production."
Simile aims to address this issue. The team trains generative agents representing real population preferences by collecting data from in-depth human interviews, historical transaction records, and texts from behavioral science journals. Users can interact with these simulated individuals or groups through simple text or image inputs and observe their reactions in specific scenarios.
Simile claims its platform can currently simulate the attitudes and behaviors of over 1,000 real individuals with an accuracy rate approaching that of humans repeatedly answering questionnaires. The technology has already demonstrated commercial value. Simile revealed that US pharmacy giant CVS Health Corp. is testing its service for inventory selection and display decisions, and Australian telecom company Telstra is an early adopter. Furthermore, after using Simile, top US automated investment service provider Wealthfront expanded its user research scope by 15 times.
Beyond market research, Simile shows potential in high-stakes business communications. The team claims its model can help public company executives rehearse earnings calls by analyzing past call transcripts and research reports, accurately predicting potential analyst questions, and even simulating the potential impact of different answers on stock prices. Simile CEO Joon Sung Park disclosed that in simulated earnings calls, the system achieved an 80% accuracy rate in predicting questions.
In a broader vision, Simile hopes to build a simulation system encompassing trillions of interactive decisions between individuals, organizations, cultures, and nations, helping users gain strategic insights into scenarios like policy changes and litigation outcome predictions.
Simile's massive funding is not an isolated case. Statistics show that in the first few weeks of 2026 alone, 17 US AI companies secured over $100 million in funding, spanning areas from foundational models to vertical applications. As agent tools like OpenClaw and Manus gain popularity in open-source communities and enterprises, AI is accelerating its evolution from "generating content" to "simulating behavior" and "taking over tasks."
Industry analysis suggests the "social simulation" sector represented by Simile signifies AI technology beginning to touch the core of human decision-making and group behavior. If this technological path is validated, traditional methods for market research, policy testing, and even corporate strategy formulation could be reshaped. In an era of increasing uncertainty, humanity might indeed need a mirror to glimpse the future, and Simile is attempting to load that mirror.