NeoCognition Raises $40M to Build Self-Learning AI Agents

NeoCognition, an AI research lab founded by Ohio State University professor Yu Su, has emerged from stealth with $40 million in seed funding. The startup aims to move beyond the current generation of unreliable AI agents by developing systems capable of autonomous, human-like specialization across any domain.
The funding round was co-led by Cambium Capital and Walden Catalyst Ventures. Additional participants included Vista Equity Partners, as well as angel investors such as Intel CEO Lip-Bu Tan and Databricks co-founder Ion Stoica.

Solving the Reliability Gap

Yu Su, who leads the lab, notes that current AI agents function primarily as generalists that require a "leap of faith" from users. According to Su, existing tools from providers like Claude Code, OpenClaw, and Perplexity successfully complete tasks as intended only about 50% of the time. Because of this lack of consistency, he argues that current agents are not yet ready to operate as trusted, independent workers.
NeoCognition seeks to address this by creating systems that can self-learn to become experts in specific fields. Su believes that while human intelligence is broad, its true power lies in the ability to rapidly master the unique rules, relationships, and consequences of new environments. By mirroring this process, NeoCognition aims to build agents that can autonomously construct a "world model" for any given profession or environment.

Enterprise-Focused Growth

NeoCognition plans to target the enterprise market, offering its agent systems to established SaaS companies. These organizations could use the technology to build autonomous agent workers or to enhance their existing product offerings. Su highlighted the strategic value of the investment from Vista Equity Partners, noting that the firm’s extensive portfolio of software companies provides a direct path to businesses looking to modernize with AI.
The startup currently employs approximately 15 people, with the majority holding PhDs. Su, who initially resisted pressure from venture capitalists to commercialize his academic research, decided to spin out the startup after observing that advancements in foundational models could finally enable truly personalized agent behavior. Unlike other approaches that require custom engineering for every specific vertical, NeoCognition is focused on building generalist agents that possess the inherent capacity for self-learning and specialization.