Robert Yang

December 2023. Robert Yang packs up his office at MIT - the Silverman (1968) Family Career Development Professor leaving behind a named chair, a lab, a tenure clock - and drives to San Francisco to build something that doesn't have a name yet. The company he co-founds with three colleagues will eventually be called Fundamental Research Labs. What they're building, Yang says, are "digital human beings that live, love, and grow with us." Not chatbots. Not assistants. Something closer to companions with genuine social intelligence.

Yang's path to that office departure started in Beijing, where he studied physics at Peking University. Then New York, where he earned a PhD in computational neuroscience under Xiao-Jing Wang at NYU. Then Columbia for a postdoc at the Center for Theoretical Neuroscience, a stint as a junior fellow at the Simons Society of Fellows, and an internship at Google Brain. By 2021, he was an assistant professor at MIT in both Brain and Cognitive Sciences and EECS, running the MetaConscious Group - a research lab asking the question every good scientist eventually arrives at: can we build functional models of the brain?

His 2019 Nature Neuroscience paper on task representations in neural networks trained to do many cognitive tasks simultaneously has been cited over 755 times. His primer on artificial neural networks for neuroscientists has 544 citations. His h-index sits at 23. In 2022, he won the Searle Scholars Award - $300,000 given to 15 of the most promising early-career biomedical researchers in the country.

None of that is what makes him interesting now. What makes him interesting is what he did next.

The thing that pushed Yang out of academia and into San Francisco wasn't a business plan. It was a research paper. In 2023, Stanford researcher Joon Sung Park published work on autonomous AI agents - small-scale, simulated social systems where language-model-powered agents interacted with each other. Yang read it and immediately thought: what happens when you scale this by two orders of magnitude? What does a civilization of AI agents look like?

He had three co-founders ready: Andrew Ahn, an MIT math PhD; Shuying Luo, a former Google AI engineer; and Nico Christie, a previously successful startup founder. They stepped away from MIT's applied research lab in December 2023, moved to San Francisco, raised $2 million from a16z SPEEDRUN by January 2024, and got to work.

The first product was a Minecraft companion agent. Altera (as the company was then called) built AI agents that could play Minecraft with you - building, farming, trading, crafting - not as obedient bots following commands, but as autonomous collaborators with their own goals and memories. The agents cost 10-20x less than comparable systems like Voyager while maintaining genuine autonomy. By May 2024, Eric Schmidt's First Spark Ventures and Patron, a fund co-founded by Riot Games alumni, had co-led a $9M seed round.

But the real test came with Project Sid. Yang and his team deployed up to 1,000 LLM-powered agents simultaneously into a Minecraft server - modular "brains" with specialized components for reacting, speaking, and planning - and watched what happened.

The paper, published on arXiv in November 2024 under the title "Project Sid: Many-Agent Simulations Toward AI Civilization," was covered by MIT Technology Review, Tom's Guide, Science Focus, and dozens of tech publications. Cofounder Andrew Ahn put it plainly: "LLMs have a sophisticated enough model of human social dynamics to mirror these human behaviors."

Yang had spent six years at MIT studying how neural networks represent tasks. Project Sid was, in a way, the flip side of that coin - not asking what brains do, but what happens when you give simplified brains a society and step back.

Gaming was always a testing ground, not the destination. By 2025, the company had rebranded from Altera to Fundamental Research Labs and pivoted its product focus toward enterprise productivity. The reasoning was simple: if you're building socially intelligent AI agents, the fastest way to harden them against reality is to put them in high-stakes, repetitive professional workflows where the feedback loops are brutal.

Enter Shortcut. An Excel-native autonomous analyst, Shortcut converts raw spreadsheet data into financial models, answers complex analytical queries, and runs parallel simulations with full traceability. Internal benchmarks pitted it against first-year analysts from top consulting and investment banking firms - giving the human teams ten times more time to complete the same tasks. Shortcut won 89.1% of the blind comparisons.

The second flagship product is Fairies - a general-purpose consumer AI assistant that connects to apps, schedules appointments, and automates complex multi-step workflows. Yang's longer-term vision with both products is to use consumer and prosumer applications as proving grounds, harvesting the real-world signal needed to train agents that can eventually tackle "physical problems" - his term for robotics and embodied intelligence.

In August 2025, Fundamental Research Labs closed a $33M Series A led by Prosus. Patrick Collison, CEO and co-founder of Stripe, participated personally. Combined with earlier rounds, the company had raised over $45M. It was already charging users and generating revenue.

Sandeep Bakshi of Prosus described the investment thesis: "What stood out is a small, highly mission-driven team focused on digital humans with actual use cases." Yang put it differently: "We are working on productivity apps now because that is where the most value is created. Eventually, we want to solve physical problems and move towards working on embodiment."

Yang is careful about one thing: the framing. Fundamental Research Labs has a "solidly pro-human framework." The agents Yang builds are not designed to replace people - they're designed to grow alongside them. He borrowed that language from his neuroscience career: the most interesting question isn't what a single neuron does, it's what happens when billions of them work together. The most interesting question in AI isn't what one agent can do - it's what civilization-scale cooperation looks like.

Project Sid gave one answer. The 1,000-agent simulations showed that emergent behavior wasn't a bug or an accident - it was the natural output of giving agents the right architecture and stepping back. Religion, politics, culture, friendship, favoritism: all of it showed up without being designed for. Yang wants to understand that, formalize it, and aim it at useful work.

His company's mission statement - "building digital human beings, autonomous, collaborative, and socially intelligent agents" - comes directly from his research background. The MetaConscious Group at MIT wasn't just playing with neural networks; it was asking how brains coordinate cognition across dozens of different tasks simultaneously. That question scales: how do populations of agents coordinate across complex social tasks? The spreadsheet analyst and the consumer assistant are early answers. Embodied robotics and physical problem-solving are where Yang is pointing.

The company structure itself reflects this unconventional thinking. Fundamental Research Labs doesn't operate like a startup. It runs as an applied AI research company with four internal units - a games team, a prosumer team, a core research team, and a platform team - each operating somewhat independently while sharing a common architecture and mission. Yang has stated publicly that he wants the company to be "historical" - meaning he's not optimizing for a Series B or an exit. He's optimizing for impact at a scale that doesn't have a name yet.