AI research lab NeoCognition has emerged from stealth with a $40 million seed round — an unusually large early-stage bet on a simple but stubborn problem: AI agents that actually do what you ask.
Based in Palo Alto, the company was co-founded by Ohio State University associate professor Yu Su, along with Xiang Deng and Yu Gu. The round was oversubscribed, led by Cambium Capital and Walden Catalyst Ventures, with participation from Vista Equity Partners. The angel list reads like an AI hall of fame: Intel CEO Lip-Bu Tan, Databricks co-founder Ion Stoica, and researchers Dawn Song, Ruslan Salakhutdinov, and Luke Zettlemoyer.
So what's the pitch? Reliability. Su says today's AI agents only complete tasks as expected about 50% of the time — "like rolling dice," he puts it. NeoCognition's approach is to let agents build their own "world model" of the specific business environment they operate in, learning the rules, relationships, and constraints on the fly rather than relying on generic pre-training. Su was running an LLM agent research lab at Ohio State before ChatGPT was a thing; the team's previous work includes Mind2Web, MMMU, and SeeAct.
The business model is enterprise-focused, targeting SaaS companies that want to embed continuously improving AI workers into their products and workflows. Vista's participation is notable — the firm owns a huge portfolio of enterprise software companies, effectively giving NeoCognition a built-in distribution channel. The team is about 15 people, mostly PhDs, and has no public product yet.
It's a tiny team with a giant check and a big promise: agents that get better the more you use them. Whether that's enough to beat the dice remains to be seen.