Breaking
Simile raises $100M Series A Index Ventures leads round backed by Fei-Fei Li & Andrej Karpathy AI predicted 8 out of 10 earnings call questions CVS Health and Telstra sign as first customers Stanford PhD becomes $100M CEO at 30-person startup Generative Agents paper: 18,000+ citations and counting Simile raises $100M Series A Index Ventures leads round backed by Fei-Fei Li & Andrej Karpathy AI predicted 8 out of 10 earnings call questions CVS Health and Telstra sign as first customers Stanford PhD becomes $100M CEO at 30-person startup Generative Agents paper: 18,000+ citations and counting
Joon Sung Park speaking at TED AI 2023

Joon Sung Park at TED AI 2023 • "A simulation of human reality, powered by AI"

Profile • AI Founder

Joon
Sung
Park

The oil painter who decided to simulate all 8 billion humans - and raised $100M to do it.

CEO, Simile Stanford PhD '25 Palo Alto, CA Series A - $100M
$100M Raised
18K+ Citations
8B Humans to Simulate
$100M Series A funding
8/10 Earnings Q's predicted
30 Team members
2 Best Paper Awards
1K People interviewed 2hrs each

Portrait of a Simulator

Before Joon Sung Park started building software that predicts how you'll react to a price change, he spent years painting portraits of ordinary objects - in hyper-realist oil, capturing the moment just before something vanishes.

That eye for transience - the way a thing changes just before you notice it changing - runs through everything he's built since. Simile, the Palo Alto AI company he co-founded and leads as CEO, does not predict the future. It replicates the present: the specific texture of how a specific person decides, hesitates, and then acts. The difference matters enormously, and it's why CVS Health has been a customer for months and why Telstra, Australia's largest mobile carrier, signed on before most people had heard of the company.

In a recent earnings call, we simulated - we actually predicted - eight out of 10 questions that were actually asked on this call.

- Joon Sung Park, CEO of Simile, Bloomberg TV interview

The mechanics: Simile trains AI agents on chat-style interviews with real people. Two-hour conversations about decisions, trade-offs, preferences. The agents absorb transaction histories, behavioral science literature, the particular cadence of how a person reasons about a purchase. At some point they stop being a model and become - in Park's own framing - a "digital twin" or "digital clone" of their human counterpart. CVS fed its system data from real customers and used the resulting AI population to test pet medicine placements and store layout decisions, running unlimited rounds of simulated polling without the fatigue that real survey respondents accumulate after the third questionnaire.

The company spent seven months doing exactly this kind of foundational work before Simile emerged from stealth in February 2026 - a deliberate slowness from a team that had every incentive to rush. Park's co-founders include Percy Liang and Michael Bernstein, two of Stanford's most prominent AI researchers, along with Lainie Yallen. The institutional backing is similarly serious: Index Ventures led the $100M Series A, with Bain Capital Ventures and Hanabi Capital joining in, plus personal checks from Fei-Fei Li and Andrej Karpathy.

What Simile actually does: Companies ask questions they used to answer with focus groups. Simile replaces the group with a simulated population - AI agents built from real-person interviews and behavioral data - that can be questioned indefinitely, at any time, about any scenario. No respondent fatigue. No sampling limitations. Early use cases include earnings call prep, product launch testing, and UI change reactions.


The Paper That Started It All

In April 2023, Park published a paper called "Generative Agents: Interactive Simulacra of Human Behavior" with co-authors Joseph O'Brien, Carrie Cai, Meredith Ringel Morris, Percy Liang, and Michael Bernstein. The concept: create software agents that don't just respond to prompts but live inside a persistent simulation - waking up, making breakfast, planning their day, running into each other, forming opinions, and remembering what happened yesterday.

To demonstrate, Park and his team built Smallville: a simulated town of 25 AI agents. They gave one agent a single instruction: throw a Valentine's Day party. What happened next is the detail that made the paper land hard. Without further prompting, the agents spread word of the party. They asked each other out on dates. They coordinated arrival times. They showed up together. The researchers hadn't programmed any of this - it emerged from agents that had memory, reflection, and planning. The paper won Best Paper at UIST 2023 and has since accumulated over 18,000 citations. Game designers started reaching out. Researchers in social policy saw it as a tool for testing assumptions before deploying interventions on real populations.

Human societies are complex systems. Our institutions, platforms, and policies rest on assumptions about how people perceive, decide, and act.

- Joon Sung Park, joonsungpark.com

The link between Smallville and Simile is direct. Smallville proved that AI agents with the right architecture could model social behavior at the group level. Simile asks: what if you grounded those agents in real people's data? What if instead of fictional characters in a simulated town, you had actual human beings - their preferences, their transaction histories, their specific reasoning styles - encoded into AI agents and then assembled into populations you could query? That shift from fictional to grounded is the core technical bet, and it's one Park has been building toward since at least 2021.


From Swarthmore to Stanford to Simulation

Park did his undergraduate work in computer science at Swarthmore College, a small liberal arts school in Pennsylvania that has produced an unusual number of people who think carefully about the intersection of technology and humanistic concerns. He went to the University of Illinois at Urbana-Champaign for his master's, where he worked under Professor Karrie Karahalios on how response time shapes human-algorithm interaction - an early signal of the kind of problem he finds interesting: not just what AI does, but how humans and AI systems perceive and respond to each other in time.

Stanford followed, where he joined both the Human-Computer Interaction and Natural Language Processing labs. He co-coined the term "foundation model" during this period - now a standard phrase across the entire AI industry. He received the Microsoft Research Ph.D. Fellowship in 2022, the Terry Winograd Fellowship in 2021, and the Siebel Scholar Award in 2019. He interned at Microsoft Research in Seattle, working with Drs. Meredith Ringel Morris and Ece Kamar. His work has been covered in The New York Times, The New Yorker, Forbes, WIRED, Nature, and Science.

He also taught CS 222 at Stanford in Fall 2024: AI Agents and Simulations. A PhD student designing and teaching a graduate course in the subject he invented. He finished his doctorate in 2025 and moved full-time to Simile.


The Painter's Eye

Hyper-realist oil painting trains you to look at something longer than is comfortable. The discipline - finding every gradation of shadow on a ceramic mug, every capillary of color in a piece of fruit - is an exercise in not simplifying. Most people look at a thing and accept a category. Hyper-realists look until the category dissolves and the specific thing remains.

Park trained extensively in this mode, focusing on transience - the moment before decay, the second just before a thing changes. It is not a coincidence that his AI research centers on capturing the specific texture of individual human decision-making rather than averaging across populations. His digital twins are not demographic profiles. They are not archetypes. They are attempts at specificity. The oil painter's instinct for the particular detail runs through the scientist's insistence on grounding agents in individual interview data rather than aggregate behavioral assumptions.

Index Ventures noticed. Their write-up on leading Simile's Series A describes Park as "an oil painter turned entrepreneur who combines creativity with operational excellence, ambition with realism, and competitiveness with empathy." Those paired contradictions are not marketing language. They are the load-bearing structure of how Simile works: ambitious enough to want to simulate all 8 billion humans, grounded enough to spend seven months interviewing a thousand of them before building anything.

From Swarthmore to $100M

2013 - 2017
B.A. in Computer Science at Swarthmore College; develops extensive background in oil painting trained in hyper-realism
2018 - 2020
M.S. in Computer Science at UIUC; thesis on response time in human-algorithm interaction under Prof. Karrie Karahalios. Siebel Scholar Award (2019)
2020
Joins Stanford's HCI and NLP labs for PhD. Terry Winograd Fellowship (2021)
2022
Microsoft Research PhD Fellowship. Internship at Microsoft Research Seattle with Meredith Ringel Morris and Ece Kamar. Best Paper Award at CHI 2022
April 2023
Publishes "Generative Agents: Interactive Simulacra of Human Behavior" - the paper that introduces the Smallville simulation and the concept of AI agents with memory, reflection, and planning
October 2023
Best Paper Award at UIST 2023 for Generative Agents. TED talk at TEDAI 2023: "A simulation of human reality, powered by AI"
Fall 2024
Teaches CS 222 - AI Agents and Simulations at Stanford as course instructor
2025
Completes PhD at Stanford. Co-founds Simile in Palo Alto with Percy Liang, Michael Bernstein, and Lainie Yallen
February 2026
Simile exits stealth. Announces $100M Series A led by Index Ventures, with participation from Bain Capital Ventures, Hanabi Capital, Fei-Fei Li, and Andrej Karpathy. Customers include CVS Health and Telstra

The Simulation Stack

Simile's technology has three layers. Understanding all three is how you understand why it's hard to replicate and why the team spent seven months on it before showing it to anyone.

🎤

Layer 1: The Interview

Two-hour one-on-one conversations with real people about how they make decisions - what they buy, why they switch brands, what makes them hesitate. This is the raw material. Park's team has conducted over 1,000 of these.

🧠

Layer 2: The Agent

Each interview becomes an AI agent that embeds the person's reasoning patterns, preferences, and behavioral signatures. Add historical transaction data and behavioral science literature. At this point, the agent is a functioning digital twin.

🌎

Layer 3: The Simulation

Assemble agents into representative populations. Run any scenario - a price change, a new product, an earnings call question - and watch how the population responds. Unlimited iterations. No fatigue. Instant feedback.

TED AI 2023 - A Simulation of Human Reality

Park gave this talk in October 2023, months before Simile existed publicly. Watch how he frames the core idea - and notice how precisely it maps to what Simile built.

► Bloomberg: AI Startup Aims to Predict Human Behavior (Feb 2026) • TED Talk: A simulation of human reality, powered by AI

Quotes

"AI agents are trained on chat-style interviews with actual people, at which point the agents become 'digital twins' or 'digital clones' of their human counterparts."

Wall Street Journal

"Simile is a real combination of amazing frontier researchers but also amazing product and engineering talent."

On Simile's Team

"Human societies are complex systems. Our institutions, platforms, and policies rest on assumptions about how people perceive, decide, and act."

joonsungpark.com

Six Things Worth Knowing

01

His name "Joon" rhymes with June. He notes this on his personal website - a small act of consideration for anyone who has to say it out loud.

02

He co-coined the phrase "foundation model" during his Stanford PhD - a term now used constantly across the entire AI industry to describe large pre-trained models.

03

His Generative Agents paper has over 18,000 Google Scholar citations. That's more than most researchers accumulate across a full career.

04

Simile's name works on two levels: it's a literary device (a comparison) and a reference to the company's core act - simulating human behavior through similarity to real individuals.

05

The Smallville experiment started with one instruction to one AI agent: throw a Valentine's Day party. The 25-agent town organized, invited, dated, and arrived - all without further prompting.

06

Simile has 30 employees and $100M in the bank. The average Silicon Valley startup raises that amount with 10x the headcount. Park and his co-founders were deliberate about hiring slowly.

Achievements

  • Best Paper Award at UIST 2023 for "Generative Agents: Interactive Simulacra of Human Behavior"
  • Best Paper Award at CHI 2022
  • Microsoft Research Ph.D. Fellowship (2022)
  • Terry Winograd Fellowship (2021)
  • Siebel Scholar Award (2019)
  • Co-coined the term "foundation model"
  • 18,000+ Google Scholar citations
  • Raised $100M Series A for Simile with backing from Fei-Fei Li and Andrej Karpathy
  • Taught CS 222 at Stanford: AI Agents and Simulations
  • TED AI 2023 speaker
  • Covered in NYT, New Yorker, Forbes, WIRED, Nature, Science, NBC News
  • Led Simile's model to predict 8 of 10 actual analyst questions at a live earnings call

Academic Path

2020 - 2025
Stanford University
Ph.D. in Computer Science - HCI & NLP Labs
2018 - 2020
UIUC
M.S. in Computer Science
2013 - 2017
Swarthmore College
B.A. in Computer Science

Know someone who should read this?