Long-horizon agents

AI systems capable of executing complex, multi-step workflows across hours or days without human intervention.

Long-horizon agents represent a shift from simple chat interfaces to autonomous operators. By utilizing advanced reasoning models like OpenAI's o1 or DeepMind's AlphaProof, these agents maintain state across hundreds of sequential actions to solve open-ended problems. They excel in high-stakes environments: a software agent might spend four hours debugging a distributed system, while a research agent could synthesize 50 separate technical papers into a single coherent report. These systems use hierarchical planning and self-correction loops to navigate 'lost in the middle' context windows, ensuring that the final output aligns with the initial objective despite thousands of intermediate tokens.

https://openai.com/index/learning-to-reason-with-llms/