Introducing Agent Memory on Databricks
Databricks Agent Memory lets agents retain context across sessions, tripling task completion while cutting tokens 31%
“agents with memory perform three times as many tasks while spending 31% fewer tokens”
Databricks launched Agent Memory in beta, a three-part system (guided sessions, managed memory, memory pipelines) built on LakeBase that gives AI agents persistent context across conversations. Internal research shows memory-enabled agents complete 3x more tasks while using 31% fewer tokens. This is a meaningful infrastructure-layer signal as the industry shifts from stateless to stateful agent architectures.