New paper from Microsoft on compressing agent context at test time.

If your agent gets worse as its interaction history grows, this method is worth a look.

FOCUS asks which past interactions the agent's next decisions actually depend on. It keeps those units of the history and drops the others. It needs no training data or fine-tuning, so it works as a separate layer in front of closed-API models.

On tool-calling, QA, web and multi-turn dialogue benchmarks, it cuts peak context by up to 48% and raises task success by up to 8.9 points compared with running on the full history.

Paper: https://academy.dair.ai/papers/focus-training-free-decision-preserving-context-compression-for-llm-agents-2609.37590