Databricks Lakebase makes Postgres ready for agents

Agentic AI is converging on Postgres but agent workloads break what classic Postgres was built to support. Each agent wants its own copy of the data, and agents create and discard those copies constantly. Demand arrives in bursts, and the work has to run on live operational data. To test whether managed Postgres changes the equation, Nucleus interviewed Databricks customers running Lakebase, a Postgres database that puts storage on the data lake. Removing data synchronization eliminated copy pipelines and reduced pipeline processing times by 70 percent. Zero-copy branching lets AI agents validate changes on instant clones and merge only approved results back to production. As a result, agents moved into billing, collections, and pricing with a solid data foundation. As agents increasingly own the operational layer built above the system of record, the Postgres services rebuilt around them will aggregate the workloads classic databases and copy pipelines cannot hold.