NeoCognition Raises $40M to Build AI Agents That Learn on the Job

NeoCognition emerged from stealth on April 21 with a $40 million seed round and a specific claim: it can build AI agents that learn and specialize on the job, the way human experts do over a career.

The company was founded by researchers Yu Su, Xiang Deng, and Yu Gu, all with backgrounds in natural language processing and AI agents. The round was co-led by Cambium Capital and Walden Catalyst Ventures.

The core pitch is generalist agents that continuously adapt to enterprise environments, narrowing their focus as they gain experience rather than staying fixed at whatever capability the base model was trained with. That's a real distinction from most agent deployments today, where models are essentially static between fine-tuning runs.

Whether it works is another matter. Continuous learning in AI systems is difficult to get right without catastrophic forgetting, where adapting to new information degrades performance on older tasks. NeoCognition hasn't published results yet. The $40 million will fund the attempt to find out.

The announcement was reported by TechCrunch.