The competition in the field of artificial intelligence is shifting from large models to more capable agents. Startup NeoCognition recently emerged from stealth mode and successfully completed a $40 million seed round, aiming to address the current shortcomings of AI agents in handling complex tasks with reliability.

This round was led by Cambium Capital and Walden Catalyst Ventures, with participation from well-known investors such as Intel CEO Pat Gelsinger. The funds will be used to accelerate the development of agent systems that can build "world models" through self-learning, similar to humans, and quickly adapt to different specialized fields.

Move beyond "hit-or-miss" operations and improve task success rates for agents

Currently, the success rate of mainstream AI agents in performing specific tasks is approximately 50%, and users often need to try repeatedly to achieve desired results. Professor Su Yu, founder of NeoCognition, pointed out that existing agents are mostly "generalists," lacking the ability to continuously learn when entering specific vertical industries.

To break through this bottleneck, NeoCognition's system mimics the process of human specialization. This agent can autonomously learn the rules, logic, and causal relationships within specific environments, rapidly becoming an "expert" in areas such as law, finance, or engineering, significantly improving the accuracy of task execution.

Focusing on the enterprise market, collaborating with software giants to drive intelligent transformation

Differing from consumer-facing products, NeoCognition plans to primarily sell its agent systems to large enterprises and software service providers. Through this B2B model, companies can use this technology to build customized "AI employees" or integrate it into existing products to enhance functionality.

Notably, the involvement of Vista Equity Partners provides the company with access to a vast customer base of software portfolios. Currently, the NeoCognition team consists of about 15 people, most of whom hold PhDs, and they are working to bring this highly specialized AI agent to a broader commercial application scenario.