Clearing chat history helps AI agents find better solutions

When an AI agent works on a long task, its growing chat history can trap it near a mediocre solution; Microsoft researchers call these regions idea basins. Their fix, Fork-and-Flush, splits the agent into parallel branches that keep the workspace files but start fresh conversations, then keeps the best result. On 13 tasks it scored 66 percent higher than single runs on the same compute budget.

Summary written with AI by HowTechMoves, based on reporting by The Decoder.