Clearing chat history helps AI agents find better solutions

Microsoft researchers discovered that autonomous AI agents often struggle to improve results because long chat histories trap them in repetitive solution patterns. To address this, they developed a technique called Fork-and-Flush that clears the conversation log while preserving the agent's current workspace. This method allows models to overcome these cognitive loops, resulting in performance gains of up to 66 percent in task execution.