Microsoft Research: Clearing conversation history can help AI agents find better solutions

A study by Microsoft Research found that long-running research AI agents can get stuck in local solution regions due to accumulated conversation context, and increasing computational power is hard to break through. The researchers proposed the Fork-and-Flush method: the agent splits into K parallel branches, each branch clears the conversation context but retains the workspace and generated files, and finally keeps the best branch. On 13 long-horizon tasks, this method achieved a normalized average score of 0.78 under the same computational budget, compared to 0.47 for single runs and 0.54 for best-of-4, with relative improvements of 66.0% and 44.4%, and achieved the highest average score on 12 of the 13 tasks; clearing context without branching improved by about 22%. The agents used GPT-5.6-terra and GPT-5.5, running via the GitHub Copilot CLI, with single runs ranging from a few hours to over 100 hours.