Google Cloud AI Research Introduces ReasoningBank: A Memory Framework that Distills Reasoning Strategies from Agent Successes and Failures

Researchers introduced ReasoningBank, a closed-loop memory framework for AI agents that retrieves relevant past lessons before a task and extracts structured strategies afterward.

Unlike systems that only store success cases, ReasoningBank learns from both successful and failed trajectories to improve future reasoning and workflow memory.

The team also paired it with memory-aware test-time scaling, using multiple attempts as training signals to strengthen memory consolidation.

The approach could make AI agents more reliable on repeated tasks by preserving and reusing lessons instead of relearning from scratch.