Inside Simile, the $2B Startup Simulating Human Behaviour

Key points

Key takeaways from Sourcery interview with Simile co-founder and CEO Joon Sung Park (October 2026):

The simulation thesis. Park argues that civilisations have an innate drive to recreate themselves, and the same instinct now backs a company: Simile builds foundation models of human behaviour, creating digital twins of real people to simulate how individuals, consumers and entire populations make decisions.

Smallville origins. The company grew out of Park's 2023 Stanford research: a virtual town of 25 generative agents with memory, planning and reflection that produced unscripted, emergent social behaviour — including a Valentine's Day party no agent had been programmed to arrange.

85% accuracy. Simile recruited a representative sample of 1,000 Americans, ran structured two-week voice interviews and trained agent twins of each participant; at the end of 2024 the agents predicted the participants' survey and behavioural responses with 85%+ accuracy — a first for the field.

Product. The platform replaces traditional market research: surveys, concept tests and conjoint and maxdiff analyses that could cost hundreds of thousands to millions of dollars over years now run almost instantly in simulation.

The science. Stated preferences follow normal distributions while real behaviour follows a steeper activation curve; optionality and opportunity cost govern decisions — so Simile models the environment and alternatives around a person, not just the person.

Scale and customers. Partnerships give access to more than 40 million real people, and clients including CVS, Wealthfront, Deloitte and Banco Itaú use simulations to understand customers, test products and optimise messaging before deploying in the real world.

Funding. Simile has raised $300M in total at a reported $2B valuation, backed by Greenoaks and Index Ventures plus AI researchers including Fei-Fei Li and Andrej Karpathy; the episode was recorded at Index Ventures' San Francisco office.

Read more: Sourcery on YouTube · Simile