NeoCognition Emerges from Stealth with $40M Seed to Build Self-Learning AI Agents That Rival Human Experts

US-based AI startup NeoCognition has exited stealth mode after securing a substantial $40 million seed funding round.

The oversubscribed round was co-led by Cambium Capital and Walden Catalyst Ventures, with participation from Vista Equity Partners and notable angels including Intel CEO Lip-Bu Tan.

Founders' Pioneering Legacy

Founded in 2025 by Ohio State University professor Yu Su, alongside Xiang Deng and Yu Gu, NeoCognition draws from a team that pioneered key AI agent benchmarks like Mind2Web, MMMU, and SeeAct.

These contributions underpin modern agents in leading models from OpenAI, Anthropic, and Google, establishing the founders as leaders in the field.

Revolutionary Self-Learning Technology

NeoCognition's core innovation lies in AI agents that self-learn on the job, autonomously building world models to specialize as domain experts much like humans.

This addresses the unreliability of current agents, which succeed only about 50% of the time, by enabling rapid mastery of unique workflows and constraints.

Enterprise Impact and Future Roadmap

Headquartered in Palo Alto, the 15-person team, mostly PhDs, plans to deploy these agents to enterprises and SaaS firms for reliable, cost-effective AI workers in high-stakes applications.

CEO Yu Su highlights that this human-mimicking specialization eliminates manual customization, promising safer and more efficient AI deployment.

Investors like Databricks co-founder Ion Stoica praise the approach for achieving the reliability needed to scale expert intelligence across industries.

As AI agents evolve from generalists to specialists, NeoCognition positions itself to democratize expertise and drive transformative productivity gains.