Simile

Simile is an American artificial intelligence company headquartered in San Francisco and Palo Alto, focused on building AI simulations of human behavior for market research, brand testing, and policy analysis. The company was founded in 2025 by Joon Sung Park, Michael Bernstein, Percy Liang, and Lainie Yallen, all with academic ties to Stanford University. Simile applies the generative-agents research lineage that originated with Park's 2023 Smallville paper to commercial AI digital-twin applications. As of May 2026, Simile is one of the principal commercial agent-simulation insurgent labs, with a $100 million Series A in February 2026 led by Index Ventures at a $1 billion-class valuation, and angel participation from Fei-Fei Li and Andrej Karpathy.

At a glance

• Founded: 2025 in San Francisco and Palo Alto by Joon Sung Park, Michael Bernstein, Percy Liang, and Lainie Yallen.
• Status: Private. Series A in February 2026 at an unconfirmed but unicorn-class valuation, with the round size at $100 million.
• Funding: $100 million Series A (February 2026) led by Index Ventures, with Bain Capital Ventures, A*, and Hanabi Capital participating. Angel investors include Fei-Fei Li and Andrej Karpathy.
• CEO: Joon Sung Park, Co-Founder and Chief Executive Officer. Stanford computer-science PhD; lead author of the 2023 Generative Agents paper.
• Other notable leadership: Michael Bernstein, Co-Founder and Chief Product Officer. Stanford computer-science professor and senior fellow at Stanford HAI. Percy Liang, Co-Founder and Chief Scientist. Stanford computer-science professor and director of the Stanford Center for Research on Foundation Models.
• Open weights: None. Simile develops closed commercial agent-simulation technology.
• Flagship products: Simile's AI-simulation platform, which generates digital twins of real human populations for survey, A/B-testing, and forecasting applications.

Origins

Simile was founded in 2025 by Joon Sung Park, Michael Bernstein, Percy Liang, and Lainie Yallen. The company's research lineage traces back to the 2023 paper "Generative Agents: Interactive Simulacra of Human Behavior," authored by Park, Bernstein, and Liang along with Joseph O'Brien, Carrie Jun Cai, and Meredith Ringel Morris. The paper, which won Best Paper at ACM UIST 2023, demonstrated a population of 25 large-language-model-driven agents in a simulated village called Smallville exhibiting believable social behavior, including memory formation, planning, and coordinated group activity.

The Smallville result anchored the founding research thesis: that agent populations grounded in real human-interview data and conditioned on language-model reasoning can produce simulations of social behavior detailed enough to be usable for downstream forecasting and decision-support applications. The team subsequently extended the research direction to grounded-agent simulations of real human populations, conducting interviews with hundreds of individuals about their lives, decisions, and personal trade-offs, and combining the qualitative interview data with historical transaction records and behavioral-science research literature.

Park completed his Stanford PhD in computer science with the generative-agents research lineage as his thesis topic. Bernstein had been Park's PhD advisor and remained a Stanford faculty member with a senior fellowship at Stanford HAI. Liang directed the Stanford Center for Research on Foundation Models (CRFM) and was a long-standing senior figure in NLP and language-model research. Yallen joined as the fourth co-founder.

The February 2026 emergence-from-stealth and Series A announcement was the company's first major public-facing milestone. The round of $100 million was led by Index Ventures with Bain Capital Ventures, A*, and Hanabi Capital participating, and with angel participation from Fei-Fei Li and Andrej Karpathy. Industry coverage characterized the round as one of the largest emerging-startup AI raises of early 2026.

Mission and strategy

Simile's stated mission is to build the first AI simulation of society, populated by agents based on real humans. The strategic premise is that grounded agent populations, conditioned on real demographic and behavioral data, can produce simulations detailed enough for commercial decision-making across product testing, marketing, and policy applications.

The strategy combines three threads. First, the foundational research lineage extending the Generative Agents work to commercial-scale populations and applications. Second, vertical-specific commercial deployments in market research, brand testing, and product-development applications. Third, continued academic-research output through the Stanford CRFM and HAI affiliations of Liang and Bernstein.

The competitive premise reflects Simile's positioning at the intersection of frontier academic-research credentials (the Stanford lineage and the Generative Agents paper), grounded-data methodology (the human-interview corpus combined with behavioral-science research), and venture-backed commercial scale (the $100 million Series A). Industry coverage has characterized Simile as the academic-research-credentialed alternative to commercial agent-simulation peers like Aaru, with the Stanford research lineage as the principal positioning differentiator.

Distribution channels include direct enterprise sales, strategic partnerships with research and consulting firms, and continued academic-research releases extending the Generative Agents lineage.

Models and products

• Simile platform. Closed commercial agent-simulation platform that generates digital-twin populations of real humans, conditioned on interview, transactional, and behavioral-science data. The platform supports survey-style queries, A/B testing, and forecasting applications across market research, brand testing, and policy domains.
• Generative Agents research lineage. Continued academic-research output extending the 2023 Smallville Generative Agents paper, with the team's Stanford affiliations supporting publication through major academic venues.

Distribution channels include direct enterprise sales and continued academic-research releases.

Benchmarks and standing

Simile's evaluation framework focuses on simulation-fidelity metrics: the accuracy with which agent populations reproduce real-world responses across surveys, A/B tests, and forecasting tasks. Standard horizontal-LLM benchmarks (Artificial Analysis Intelligence Index, LMArena, GPQA Diamond) do not apply to the agent-simulation use case.

The Smallville and subsequent grounded-agent-simulation research lineage anchors the company's research-credibility positioning. The 2023 Generative Agents paper, which won Best Paper at ACM UIST, has been one of the most cited research papers in the broader agent-architecture literature through 2024 to 2025, and the team's continued publication track supports recruiting and customer-credibility positioning.

Direct head-to-head benchmarks against peer commercial agent-simulation companies (Aaru in particular) remain limited because no industry-standard benchmark suite has emerged. The category-relevant benchmarks generally include grounded-agent-population fidelity to real survey-response distributions, predictive accuracy on observable forecasting tasks, and the qualitative coherence of simulated agent behavior under repeated interaction.

Leadership

As of May 2026, Simile's senior leadership includes:

• Joon Sung Park, Co-Founder and Chief Executive Officer. Stanford computer-science PhD; lead author of the 2023 Generative Agents paper.
• Michael Bernstein, Co-Founder and Chief Product Officer. Stanford computer-science professor; senior fellow at Stanford HAI; co-author of ImageNet and the 2023 Generative Agents paper.
• Percy Liang, Co-Founder and Chief Scientist. Stanford computer-science professor; director of the Stanford Center for Research on Foundation Models; long-standing senior figure in NLP and language-model research.
• Lainie Yallen, Co-Founder.
• Senior research, engineering, and commercial staff across the platform and customer-engagement programs.

The Stanford academic-research connections continue to support technical-research recruiting and customer-credibility positioning, and the Bernstein and Liang faculty appointments anchor the team's Bay Area research-community presence.

Funding and backers

• Series A (February 2026): $100 million led by Index Ventures, with Bain Capital Ventures, A*, and Hanabi Capital participating. Angel participation from Fei-Fei Li, Andrej Karpathy, and other senior AI-industry figures.

The Series A composition is notable for the academic-research-affiliated angel participation, with Fei-Fei Li (founder of World Labs) and Andrej Karpathy (former senior researcher at OpenAI and Tesla) both participating. The Index Ventures lead positions Simile within the firm's Bay Area enterprise-AI portfolio.

Industry position

Simile occupies a structurally distinctive position among 2025 to 2026-vintage agent-simulation insurgent labs. The combination of Stanford research credentials, the 2023 Generative Agents paper lineage, the grounded-interview-data methodology, and the Index-led Series A produces a profile differentiated from peer agent-simulation startups oriented around commercial-traction-first positioning.

Strategic risks include the open question of whether grounded agent simulations of real human populations produce reliably better decision-support outcomes than peer commercial agent-simulation platforms, the competitive pressure from Aaru with consulting-firm reference customers and election-cycle demonstration credibility, and the early-stage commercial-traction profile that February 2026 launch coverage acknowledged.

Strategic strengths include the Stanford research lineage that anchors academic-credibility positioning, the Bernstein and Liang faculty affiliations that support recruiting and Bay Area research-community presence, the Fei-Fei Li and Andrej Karpathy angel participation that signals senior-AI-industry confidence, and the grounded-interview-data methodology that distinguishes Simile from agent-simulation approaches relying solely on demographic conditioning.

Competitive landscape

Simile competes with several agent-simulation and adjacent peers:

• Aaru. Direct agent-simulation peer with the New York-Democratic-primary forecasting result and consulting-firm partnerships. Different commercial-traction-first positioning relative to Simile's academic-research positioning.
• Traditional market-research firms. Ipsos, Kantar, Nielsen, and other established players whose human-panel methodologies the agent-simulation category positions itself against.
• Polling firms. Pew, Gallup, Marist, and other established political-polling organizations whose methodologies grounded-agent simulations are compared against.
• Scale AI. Adjacent synthetic-data and evaluation peer.
• Stanford HAI. Research peer through the team's Stanford CRFM and HAI affiliations.
• Consulting firms with internal AI-prediction practices. Accenture, McKinsey, Boston Consulting Group, and others.

Outlook

• The cadence of public-research releases extending the Generative Agents lineage.
• The translation of the grounded-interview-data methodology into commercial reference customers.
• The competitive dynamics with Aaru across academic-credibility and commercial-traction dimensions.
• The integration of behavioral-science research into agent-population conditioning across additional vertical applications.
• The trajectory of annual recurring revenue across 2026 toward institutional-grade operating-business metrics.
• The continued tension between academic-research output and closed commercial product strategy.

Sources

• Simile official site. Company reference.
• SiliconANGLE: AI digital twin startup Simile raises $100M in funding. Series A funding context.
• Tech Funding News: $100M for Stanford spinout Simile. Stanford lineage and funding context.
• Generative Agents: Interactive Simulacra of Human Behavior (arXiv). Foundational research paper.
• Joon Sung Park personal site. Founder reference.
• Stanford HAI: Computational Agents Exhibit Believable Humanlike Behavior. Research lineage context.