NeoCognition raises $40M to make agents that learn like humans
NeoCognition came out of stealth on April 21 with a $40M seed co-led by Cambium Capital and Walden Catalyst Ventures, with Vista Equity Partners in the round and a frankly absurd angel list — Intel CEO Lip-Bu Tan, Databricks co-founder Ion Stoica, plus Dawn Song, Ruslan Salakhutdinov, and Luke Zettlemoyer as advisors. That is basically half the names you would put on the top of an AI agents reading list.
The company is led by Yu Su, Sloan Research Fellow at Ohio State, who has been running one of the more established academic agent labs in the country. The pitch is specialized intelligence: build agents that do not just ship with whatever the base model knows, but keep learning and specialize in any domain to expert level. The framing is "agents that learn like humans" — meaning they are betting against the static-weights, pure-context-stuffing approach that most coding agents use today.
Why it matters: most VC-backed agent companies are wrappers, distribution plays, or vertical pickers. NeoCognition is squarely in the continual-learning lane, the same lane Naive.AI and a handful of others see as the next moat. $40M from a syndicate this strong, anchored by an academic founder, says the early-research investors think the wrapper era is closing.
TechCrunch coverage at techcrunch.com/2026/04/21/ai-research-lab-neocognition-lands-40m-seed-to-build-agents-that-learn-like-humans. Worth watching what they ship — talent density and capital are there; the question is whether the lab-to-product translation works.
← Back to all articles
The company is led by Yu Su, Sloan Research Fellow at Ohio State, who has been running one of the more established academic agent labs in the country. The pitch is specialized intelligence: build agents that do not just ship with whatever the base model knows, but keep learning and specialize in any domain to expert level. The framing is "agents that learn like humans" — meaning they are betting against the static-weights, pure-context-stuffing approach that most coding agents use today.
Why it matters: most VC-backed agent companies are wrappers, distribution plays, or vertical pickers. NeoCognition is squarely in the continual-learning lane, the same lane Naive.AI and a handful of others see as the next moat. $40M from a syndicate this strong, anchored by an academic founder, says the early-research investors think the wrapper era is closing.
TechCrunch coverage at techcrunch.com/2026/04/21/ai-research-lab-neocognition-lands-40m-seed-to-build-agents-that-learn-like-humans. Worth watching what they ship — talent density and capital are there; the question is whether the lab-to-product translation works.
Comments