Ohio State professor-led NeoCognition emerges with $40 million seed round

NeoCognition has stepped out of stealth with a $40 million seed round and a high-stakes claim: AI agents can be built to learn a specific job well enough to be trusted inside real businesses. That pitch, led publicly by Ohio State professor Yu Su, lands in a crowded market where many startups promise autonomy but far fewer can show consistent performance.

The round was co-led by Cambium Capital and Walden Catalyst Ventures, with participation from Vista Equity Partners. The company also drew support from Intel chief executive Lip-Bu Tan, Databricks co-founder Ion Stoica, and other named backers including Dawn Song, Luke Zettlemoyer and Ruslan Salakhutdinov. NeoCognition says its founding team is Yu Su, Xiang Deng and Yu Gu, and that the three collectively bring more than 30 years of research experience on AI agents.

Su’s academic and industry background gives the startup credibility, but also raises the bar for proof. He is an assistant professor in The Ohio State University Department of Computer Science and Engineering, where he co-directs the OSU NLP group and co-leads AI-related teams in the ICICLE AI Institute and Imageomics Institute. Before Ohio State, he was a senior researcher at Microsoft Semantic Machines, working on conversational AI.

The company’s central argument is that today’s AI agents are too unreliable to serve as independent workers. Su said current agents are generalists and that users still have to take a leap of faith each time they assign a task. He said they succeed only about half the time, a failure rate that is hard to square with business functions that depend on consistency, accountability and repeatable outcomes.

NeoCognition’s answer is specialization. Rather than pushing one model to act like a universal employee, the startup says it is building agents that can self-learn within a narrow domain, develop an internal world model for that micro-environment and improve over time. The company’s own launch materials frame the business around AI agents, augmented workflows, a KPI and ROI cockpit, and AI governance features, underscoring an effort to sell measurable operational gains rather than abstract technical progress.

That distinction matters because the agent market is already full of bold claims about autonomy. NeoCognition now has to define self-learning in measurable terms: what changes after deployment, how performance is tracked, how errors are caught, and what prevents a system from drifting into unsafe or unreliable behavior as it adapts. Investors may be willing to fund the vision, but enterprises will demand evidence that learning improves results without weakening control.

For now, the startup is positioning itself as infrastructure for business software, not just another model company. Its next challenge is turning the language of expert agents and business priorities into proof that self-improving AI can be specialized, governed and dependable enough for production use.