NeoCognition, an artificial intelligence research lab spun out from Ohio State University, has emerged from stealth with $40 million in seed funding to develop AI agents capable of self-learning expertise in any domain. The funding round was co-led by Cambium Capital and Walden Catalyst Ventures, with participation from Vista Equity Partners and notable angel investors including Intel CEO Lip-Bu Tan and Databricks co-founder Ion Stoica.
Founded by Yu Su, an Ohio State professor leading an AI agent lab, NeoCognition aims to address a critical limitation in current AI agent technology: inconsistency. According to Su, today’s AI agents—whether from tools like Claude Code, OpenClaw, or Perplexity’s computer use features—successfully complete tasks as intended only about 50% of the time. This unreliability prevents them from being trusted as independent workers, he told TechCrunch in an interview published April 21, 2026.
Su argues that whereas artificial intelligence systems are often designed as generalists, their true potential lies in the ability to specialize—much like human expertise. “When we enter a recent environment or profession, we can rapidly master its unique rules, relationships, and consequences,” Su said. NeoCognition’s approach focuses on building agents that can autonomously learn and adapt to specific domains through experience, mimicking how humans develop proficiency over time.
The startup’s technology is rooted in research conducted at Ohio State University, where Su leads a lab dedicated to advancing AI agent capabilities. By leveraging recent advances in foundational models, NeoCognition believes it can create systems that move beyond brittle, prompt-dependent behaviors toward robust, self-improving agents capable of sustained performance in complex, real-world environments.
Investor Confidence in Specialized AI Agents
The $40 million seed round reflects growing investor interest in startups that aim to create AI more reliable and efficient through architectural innovation rather than scaling alone. Cambium Capital and Walden Catalyst Ventures, known for backing deep-tech ventures, co-led the investment, signaling confidence in NeoCognition’s research-driven approach.
Angel investor participation further underscores industry belief in the founding team’s vision. Lip-Bu Tan, who became CEO of Intel in 2023, has a long history of investing in transformative semiconductor and computing technologies. Ion Stoica, a professor at UC Berkeley and co-founder of multiple successful infrastructure software companies including Databricks and Anyscale, brings expertise in distributed systems and AI infrastructure—areas critical to scaling agent-based systems.
Vista Equity Partners, a global investment firm focused on enterprise software, data, and technology-enabled businesses, also participated in the round. Their involvement suggests potential interest in NeoCognition’s applications for enterprise workflow automation, where consistent agent performance could significantly reduce operational friction.
Ohio State’s connection to the startup extends beyond its founder. The university’s Early Investor Network (OSEIN), which connects accredited alumni and friends to early-stage startups, reported in January 2026 that it had grown to over 120 members and invested more than $2.5 million in Buckeye-connected startups since its inception. While OSEIN’s portfolio includes companies like Coologics and smallTalk Technology, there is no public indication that it directly invested in NeoCognition as of the funding announcement.
How Self-Learning Agents Could Change AI Use
Current AI agents often fail when faced with tasks outside their narrow training or when environmental conditions shift slightly. This brittleness limits their deployment in dynamic settings such as customer service, healthcare administration, or scientific research, where adaptability is essential.
NeoCognition’s hypothesis is that by enabling agents to identify patterns, form internal models of cause and effect, and refine their behavior through feedback—similar to how humans apprentice in a new profession—these systems could achieve higher success rates over time. Such capabilities would be particularly valuable in domains requiring nuanced judgment, such as legal research, financial analysis, or technical troubleshooting.
The company has not yet disclosed specific benchmarks or timelines for when its agents might reach human-equivalent performance in specialized tasks. Even though, Su emphasized that the goal is not to replicate broad human cognition but to engineer agents that can reliably acquire and apply domain-specific knowledge through experience.
Challenges and Industry Context
Building AI agents that learn like humans remains a significant scientific challenge. While large language models have demonstrated impressive generalization, they lack persistent memory and true experiential learning without external retraining. Approaches such as reinforcement learning, neural symbolic systems, and cognitive architectures are being explored across academia and industry to bridge this gap.
NeoCognition enters a competitive landscape that includes efforts by major AI labs to improve agent reliability. For example, OpenAI has explored techniques like process supervision to improve reasoning fidelity, while Anthropic has focused on constitutional AI to enhance safety and consistency. However, few startups are explicitly framing their mission around the human-like specialization model that NeoCognition advocates.
Success will depend on the startup’s ability to translate cognitive science principles into scalable AI architectures. Key hurdles include designing reward structures that encourage meaningful exploration, preventing reward hacking, and ensuring that learned behaviors generalize safely across similar but novel situations.
What’s Next for NeoCognition
As of the April 2026 announcement, NeoCognition has not disclosed plans for a public product launch or pilot partnerships. The company stated that the seed funding will be used to expand its research team, accelerate experimentation with agent learning paradigms, and validate early prototypes in controlled environments.
Industry observers will be watching for peer-reviewed publications, conference presentations, or beta access programs that could provide insight into the technical approach. Given the founders’ academic backgrounds, This proves likely that NeoCognition will prioritize early validation through scholarly channels before pursuing commercial deployment.
For updates on NeoCognition’s progress, interested parties can follow the company’s official channels or monitor announcements from its lead investors. Cambium Capital and Walden Catalyst Ventures typically share portfolio updates through their websites and press releases when milestones are reached.
The development of AI agents capable of reliable, self-directed learning represents one of the most consequential frontiers in artificial intelligence. If successful, NeoCognition’s work could help shift AI from a tool requiring constant human oversight to a collaborative partner capable of growing expertise alongside its users.
What are your thoughts on the future of AI agents that learn like humans? Share your perspective in the comments below, and feel free to share this article with others interested in the evolution of intelligent systems.