A new AI research lab called NeoCognition has emerged from stealth with $40 million in seed funding and a bold thesis: today's AI agents are unreliable generalists that succeed only about half the time, and the solution is building agents that can teach themselves to become experts the same way humans do.
The Problem With Current Agents
NeoCognition founder Yu Su, an Ohio State professor who runs an AI agent research lab, is blunt about the current state of the technology. Whether you are using Claude Code, OpenClaw, or Perplexity's computer tools, AI agents successfully complete tasks as intended only about 50 percent of the time.
That failure rate means agents cannot yet be trusted as independent workers. Every time you ask a current agent to do something, you are taking a leap of faith. For enterprise customers who need reliability, consistency, and domain expertise, a coin-flip success rate is not acceptable.
Su argues the core issue is that today's agents are built as generalists. They know a little about everything but are experts in nothing. They lack the ability to specialize to learn the unique rules, relationships, and consequences of a specific environment and become genuinely proficient within it.
How NeoCognition Is Different
NeoCognition's approach is inspired by how human intelligence actually works. Humans have broad general knowledge, but the real power is our ability to rapidly specialize when we enter a new domain. A doctor entering a new specialty, a developer learning a new codebase, a trader understanding a new market in each case, the human builds what Su calls a "world model" for that specific environment.
NeoCognition is building agents that mirror this process. Rather than being custom-engineered for a single vertical, the agents are generalists capable of self-learning and specializing in any domain autonomously. They observe, learn the rules, and build an internal model of whatever micro-world they are placed into then use that model to operate reliably within it.
This is fundamentally different from how most AI agent companies approach the problem. Current approaches either build narrow, task-specific agents that work well in one context but cannot transfer, or deploy general-purpose agents that work inconsistently across everything.
The Funding and the Team
The $40 million seed 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.
The Vista investment is strategically significant. As one of the largest private equity firms in the software space, Vista manages a vast portfolio of enterprise SaaS companies giving NeoCognition direct access to potential customers looking to modernize their products with AI agents.
Su initially resisted VC pressure to commercialize his academic research. He finally took the leap last year when he saw that advances in foundation models had reached a point where truly personalized, self-learning agents were becoming feasible. The company currently has about 15 employees, the majority of whom hold PhDs.
The Enterprise Play
NeoCognition plans to sell its agent systems primarily to enterprises, including established SaaS companies that want to embed agent capabilities into their existing products. Rather than competing directly with consumer-facing tools like Claude Code or Codex, NeoCognition is positioning itself as the infrastructure layer that makes other companies' agents more reliable.
The pitch to enterprise buyers is straightforward: instead of deploying agents that work half the time, use NeoCognition's self-learning system to build agents that get better the longer they operate in your specific environment. For companies where agent reliability is not just a nice-to-have but a requirement healthcare, finance, legal, manufacturing this approach addresses the most fundamental barrier to AI adoption.
The Bigger Picture
NeoCognition's emergence reflects a broader shift in the AI agent market. The initial wave of excitement around agents focused on what they could theoretically do. The current wave is focused on making them actually work reliably, consistently, and in domains where mistakes have real consequences.
If NeoCognition can deliver on its promise of agents that learn like humans and specialize like experts, it could become one of the most important infrastructure companies in the AI agent ecosystem. The $40 million seed round unusually large for a company still in stealth suggests investors are betting that the self-learning approach is the key to unlocking the next phase of enterprise AI.
