AI research lab NeoCognition lands $40M seed to build agents that learn like humans
The startup aims to solve AI agents' 50% failure rate by mimicking how humans rapidly master new domains.
NeoCognition, a new AI research lab founded by Ohio State professor Yu Su, has launched from stealth with a substantial $40 million seed funding round. The investment was co-led by Cambium Capital and Walden Catalyst Ventures, with notable participation from Vista Equity Partners and angel investors including Intel CEO Lip-Bu Tan and Databricks co-founder Ion Stoica. Su, who initially resisted commercializing his academic work, was convinced to start the company by recent foundational model advances that could enable truly personalized AI agents.
Su argues that today's AI agents from companies like Anthropic (Claude Code), OpenAI, and Perplexity are unreliable generalists, successfully completing tasks only about 50% of the time. NeoCognition's core thesis is that for agents to become trustworthy, independent workers, they must be able to self-learn and specialize rapidly in any given domain, mirroring how humans master new professions. The company is building agents that can autonomously construct a "world model" of a specific environment, which Su views as the critical missing link for reliability. With a team of about 15, mostly PhDs, NeoCognition plans to sell its agent systems to enterprise SaaS companies, leveraging Vista Equity's vast portfolio for go-to-market access.
- Raised $40M seed from Cambium Capital, Walden Catalyst, and Vista Equity, with angels including Intel's CEO.
- Aims to solve the ~50% failure rate of current AI agents by enabling autonomous specialization.
- Plans to sell self-learning agent systems to enterprise SaaS companies to build AI workers.
Why It Matters
If successful, it could create reliable, autonomous AI workers for enterprises, moving beyond today's error-prone assistants.