Perplexity hires NYU professor Andrew Gordon Wilson to lead research

Why it matters

Perplexity is adding research leadership for the problems behind persistent AI agents: learning from experience, coordinating with other agents and remaining reliable as behavior changes.

Andrew Gordon Wilson (@andrewgwils) is joining Perplexity as research lead, giving the AI search company an academic specialist in generalization, probabilistic machine learning and the mechanics of building systems that keep learning after deployment.

Wilson announced the appointment in a post on X on August 24th. He said the group plans to work on continual learning and collaboration between AI agents, with research conducted "in the open." He also invited researchers interested in those areas to contact him, making the announcement an early recruiting pitch for the group he will lead.

Wilson will continue running his academic research group at New York University, where he is a professor of computer science and data science at the Courant Institute of Mathematical Sciences and the Center for Data Science. His NYU research page describes a program aimed at understanding learning and decision-making well enough to build autonomous intelligent systems, covering model generalization, uncertainty, distribution shifts, scientific discovery and the foundations of deep learning. (cims.nyu.edu)

That work gives Perplexity a research leader whose recent agenda closely matches the problems created by increasingly autonomous AI products. Wilson co-authored "When Should We Orchestrate Multiple Agents?", a 2025 paper examining when coordination among agents improves decisions, including selecting other agents and outsourcing tasks. He has also worked on epiplexity, a framework for measuring the useful structure that a computationally limited learner can extract from data. A paper published this month applied that framework to data selection and synthetic data generation, reporting that higher measured epiplexity predicted stronger transfer to new tasks. (arxiv.org)

Wilson's route to Perplexity runs through academia rather than another commercial AI laboratory. According to his curriculum vitae, he completed a doctorate in machine learning at the University of Cambridge, worked as a research fellow at Carnegie Mellon University and taught at Cornell University before joining NYU in 2019. NYU promoted him to professor in 2024. His work has included Bayesian deep learning, Gaussian processes, model generalization and open-source machine-learning software. (cims.nyu.edu)

He has also pushed against the idea that modern neural networks require entirely new theories to explain their behavior. In a 2025 ICML position paper, Wilson argued that familiar generalization frameworks can explain phenomena often treated as unusual properties of deep learning, including overparameterization and double descent. That emphasis on identifying usable principles beneath model behavior is directly relevant to agents that must adapt over long tasks without becoming less predictable. (proceedings.mlr.press)

Perplexity is building a research organization around agents

Perplexity Research lists search, reasoning, agents and systems as its core areas, and currently advertises AI researcher positions in New York City, Palo Alto and San Francisco. Wilson's invitation to prospective hires indicates that his appointment comes with a mandate to add researchers rather than serve as an academic adviser attached to an existing product group. Perplexity has not specified the size of the group Wilson will oversee or identified the first projects it plans to publish. (research.perplexity.ai)

The hire follows Perplexity's expansion from web answers into longer-running agent tasks. In July, Perplexity introduced SPACE, an internal sandbox system designed to let agents run code, preserve files and execute multi-step work over extended sessions. Perplexity said at the time that SPACE powered every session in its Computer product and had handled millions of sandbox creations during its first week of full deployment. Those figures are company-reported, but the underlying engineering direction is clear: Perplexity is preparing its products for agents that act over time, retain state and interact with more tools. (perplexity.ai)

Continual learning raises a harder research problem than adding another tool connector. A system that changes through experience must decide what to retain, how to transfer knowledge to unfamiliar tasks and how to avoid degrading abilities it already has. Multi-agent systems add questions about when agents should cooperate, delegate or rely on another agent's output. Wilson has spent his academic career studying closely related questions about uncertainty, generalization and decision-making.

His promise to publish research openly also gives Perplexity a recruiting instrument. Frontier AI laboratories compete for researchers partly through access to computing infrastructure and partly through the ability to publish work that carries academic value beyond a single product. Wilson is retaining his NYU group while taking the Perplexity role, creating a bridge between that academic incentive structure and a company deploying agent systems at product scale.

Wilson said further announcements are planned. The appointment already defines the research bet: Perplexity wants agents that can learn and coordinate after the initial model training cycle, and it has put an academic known for studying how learning systems generalize in charge of pursuing it.