A lab trained an AI system to handle research tasks that take months — by breaking them into smaller chunks

Researchers built KLong, an AI system that can work on extremely long tasks (like reading and synthesizing research papers) by splitting them into overlapping pieces during training, then practicing with progressively longer timeouts. This means AI agents can now attempt work that requires sustained focus and memory over weeks or months, not just minutes.

Until now, AI systems struggled with tasks that require hours or days of continuous reasoning — they'd lose track of context or give up. KLong shows that you can train an AI to stay on task much longer by teaching it in a structured way, then gradually pushing it to work longer. The system outperformed much larger competitors on benchmarks for research synthesis and coding, which suggests long-horizon reasoning is becoming trainable, not magical.