NeoCognition Raises $40M Seed To Build Self-Learning AI Agents With Cambium Capital And Walden Catalyst

NeoCognition, a research lab developing self-learning AI agents, emerged from stealth with a $40 million seed funding round aimed at building agents that can learn on the job to become specialized experts.

“A $40 million seed round aims to turn research on self learning agents into dependable, enterprise ready AI that can specialize without constant human oversight”

mezha.net

The funding round was co-led by Cambium Capital and Walden Catalyst Ventures, with participation from Vista Equity Partners and angels, including Intel CEO Lip-Bu Tan and Databricks co-founder Ion Stoica.

mezha.net

Yu Su, an Ohio State professor leading an AI agent lab, said he initially resisted pressure from VCs to commercialize his work before taking the leap last year to spin out the startup.

TechCrunch reported that NeoCognition is building agents that “learn like humans,” while PR Newswire said the lab’s mission is to expand access to expertise by developing AI agents that continuously learn to reach expert-level intelligence.

The round was described by PR Newswire as “oversubscribed” and said it included angel investors and founding advisors such as A&E Investments, Salience Capital Partners, Nepenthe Capital, Frontiers Capital, and leading AI researchers like Dawn Song, Ruslan Salakhutdinov, and Luke Zettlemoyer.

NeoCognition’s leadership and research pedigree were emphasized across outlets: PR Newswire named Yu Su as CEO and Co-Founder and said his team began developing large language model-based agents well before the ChatGPT moment.

TechCrunch said NeoCognition currently has about 15 employees, the majority of whom hold PhDs, and PR Newswire said the company is headquartered in Palo Alto, California.

Across the reporting, NeoCognition’s pitch is anchored in a critique of today’s agent reliability, with Yu Su arguing that current systems succeed only about half the time.

TechCrunch quoted Su saying, “Today’s agents are generalists,” and added that “Every time you ask them to do a task, you take a leap of faith.”

PR Newswire

TechCrunch reported that Su said current agents, whether from Claude Code, OpenClaw, or Perplexity’s computer tools, successfully complete tasks as intended only about 50% of the time.

The mezha.net write-up echoed the same core claim, stating that “modern agents, including Claude Code, OpenClaw and Perplexity tools, succeed at completing tasks only about half the time.”

PR Newswire framed the problem as a fundamental limitation of AI execution, quoting Su: “AI today is fundamentally unreliable when it comes to executing real work that requires deep expertise.”

TechCrunch said Su argued that because agents are still so unreliable, they are “not ready to be trusted, independent workers.”

NeoCognition’s solution, as described by TechCrunch, is to develop an agent system that can self-learn to become an expert in any domain, “similar to how humans learn.”

NeoCognition’s approach is repeatedly described as mirroring how humans build expertise by specializing through continuous learning in a “micro-world.”

“Investors are aggressively courting AI researchers to build startups that can make AI more reliable and efficient”

TechCrunch

TechCrunch said Su argued that while human intelligence is broad, its real power is “our ability to specialize,” and it quoted him explaining that “For humans, our continued learning process is essentially the process of building a world model for any profession, any environment.”

TechCrunch tied that concept to NeoCognition’s plan by quoting Su: “We believe for agents to become experts, they need to learn autonomously to build a model of any given micro world.”

PR Newswire described the lab’s research as creating agents that “continuously learn the structure, workflows, and constraints of the environments they operate in,” and it said expert agents become “faster, more cost-effective, and more reliable.”

The PR Newswire release also quoted Su saying the approach would “eliminate the extensive manual customization required by current models.”

The mezha.net article emphasized the specialization mechanism, stating that “although human intelligence has broad reach, its true strength lies in the ability to specialize quickly.”

The Tech Buzz alternative outlet contrasted NeoCognition’s adaptive learning focus with agents that “operate on pre-trained models with retrieval-augmented generation bolted on top,” arguing they “don’t actually learn from experience the way a human expert does.”

The seed round coverage foregrounded endorsements from investors and prominent AI figures, with multiple outlets quoting leaders about what NeoCognition could change.

PR Newswire quoted Lip-Bu Tan, Founding Managing Partner of Walden Catalyst Ventures, saying, “Dr. Su and his team have already developed research that spans every piece of the agent puzzle, ranging from perception to memory, planning, evaluation, and safety,” and it added, “We are confident NeoCognition is uniquely positioned to tackle the hardest challenges in agentic AI.”

The Tech Buzz

PR Newswire also quoted Ion Stoica, UC Berkeley Professor and Databricks Co-Founder, saying, “NeoCognition's new approach to building agents that learn to become experts has the potential to reach the level of reliability, efficiency, and cost-effectiveness required for high-stakes applications.”

Cambium Capital’s Landon Downs was quoted as saying, “We have strong conviction in the team's expertise and believe their research is charting a new path toward specialized intelligence that will democratize access to frontier agent capabilities.”

TechCrunch highlighted Su’s own framing of the reliability problem and the specialization thesis, quoting him on the “leap of faith” and on agents needing to learn autonomously to build a model of a micro world.

PR Newswire expanded on Su’s background by stating that his team contributed work such as Mind2Web, MMMU, and SeeAct, and it said research from their team is “widely used in every frontier large language model from OpenAI, Anthropic, and Google.”

The mezha.net article echoed the investor framing by describing the Vista Equity Partners investment as adding value through “access to a large portfolio of companies seeking to modernize their solutions with artificial intelligence.”

NeoCognition’s commercialization plan is described as enterprise-focused, with TechCrunch saying it intends to sell its agent systems primarily to enterprises, including established SaaS companies, so they can build agent workers or enhance existing product offerings.

“A $40 million seed round aims to turn research on self learning agents into dependable, enterprise ready AI that can specialize without constant human oversight”

mezha.net

TechCrunch highlighted why Vista Equity Partners’ involvement matters, saying Su pointed to an investment from Vista as especially valuable because the firm can provide direct access to a vast portfolio of companies looking to modernize products with AI.

mezha.net

The mezha.net article similarly stated that NeoCognition focuses primarily on selling agent systems to enterprises, including “well-known SaaS companies,” and it described the Vista role as enabling “access to a large portfolio of companies seeking to modernize their solutions with artificial intelligence.”

PR Newswire framed the enterprise value proposition in terms of reliability, efficiency, and safety, saying the expert agents become “faster, more cost-effective, and more reliable,” and that deeper understanding of environments enables agents to be “more responsible and safer actors in high-stake settings.”

PR Newswire also positioned the lab’s research as a new class of agents that continuously learn and specialize, with the goal of eliminating extensive manual customization required by current models.

The Tech Buzz alternative outlet added a market framing, claiming the startup is targeting enterprise applications where domain expertise traditionally requires years of human training, and it described the seed round as “one of the largest seed rounds in the AI agent space this year.”

TechCrunch emphasized that while it is possible to train agents for autonomous tasks, they must be custom-engineered for a specific vertical, and it said NeoCognition is different because it’s building generalists capable of self-learning and specializing in any domain.