CharlieKey takeaways from your favorite podcasts
← 26 October 2026Sourcery with Molly O'Shea

The $2B AI Startup Simulating Human Behavior with 85% Accuracy

Simile's founder June argues that human behavior can be simulated from grounded data on real people, predicting it with 85%+ accuracy, and that these simulations let companies test decisions before making them.

Watch on YouTube ↗

Charlie's summary

  1. Simulate individuals first, then put them in an environment and let interactions produce society

    June thinks a society simulation starts with faithful simulations of individuals, built from stable traits like risk tolerance and taste. You then place them in a shared environment, and the complex, chaotic part is “what actually gets born in their interactions.” In the 2023 Smallville experiment, 25 language-model agents with memory, planning and reflection organized a Valentine's Day party with nothing pre-programmed.

    ▶ 2:31
  2. Grounded in interviews with 1,000 real people, agent twins predicted behavior with 85%+ accuracy

    Simile's foundation came from recruiting a representative sample of 1,000 US people. An AI agent interviewed each one voice to voice, asking for their life story and hard decisions. Those interviews trained twin agents. Two weeks later, the twins predicted the people's answers on surveys and behavioral studies. At the end of 2024 this reached 85%+ accuracy. June insists simulation must be grounded and “evaluated alongside people,” not created from a vacuum.

    ▶ 9:09
  3. The gap between what people say and what they do follows different curves, and optionality drives it

    June says survey answers tend to follow a normal distribution, but actually doing something, like going to the gym, has a much higher activation energy and looks more like an exponential curve. What governs behavior is optionality: you could exercise, but you could also sleep an hour longer or take a meeting. Simile's modeling challenge is learning this translation layer from stated attitudes to fundamental tendencies. That is why it models the environment and alternatives, not just people.

    ▶ 12:16
  4. Simulation turns studies that cost hundreds of thousands or millions of dollars into a query

    Simile's model takes a description of a person or group and simulates what they would do. Insights teams use it for concept testing, conjoint and max-diff analysis, which June says could cost hundreds of thousands or millions over years and can now run almost instantly. Customers include CVS, Banco in Brazil and large financial institutions. Simile keeps vertical-specific simulated panels ready on day one and adds data for a customer's own population.

    ▶ 15:49

Give your email to unlock the rest of the key points — for every episode.

Do more with Charlie

Free — sign in with your email, no password.