Databases Enter the AI-Generated Era

From agentic AI arises the need for new OLTP architectures… such as lakebase.

In late January, Databricks published a report titled “State of AI Agents” that generously championed this premise. A few days later, it announced the general availability of its own lakebase offering.*

Beyond this alignment, the report includes a few figures based on telemetry from “more than 20,000 customers”.

The multi-LLM approach is gaining ground

The share of customers using at least three LLMs tends to rise.

Across all considered economic sectors, the share of customers using at least 3 LLMs exceeded 50% during the August–October period. The highest rate—around 65%—was in retail. The utilities sector surpassed 60%, as did healthcare, manufacturing, and financial services.

Few batch jobs, lots of real-time processing

In May and October, 96% of requests were processed in real time, with the remainder handled in batches. The technology sector shows the widest gap (32 real-time requests for 1 batch). This is followed by healthcare (13/1), likely reflecting the critical situations that organizations in that sector manage.

The creation of databases, largely agent-driven

Based on Neon telemetry—the PostgreSQL backbone at the heart of its lakebase—Databricks states that the majority of databases are now created by AI agents. Specifically, 80% in October 2025, up from 27% a year earlier. The creation of branches (cloning) followed the same trajectory (from 18% to 97%).

A pragmatic use of AI

The market watch emerges as the primary AI use within the Databricks ecosystem for the studied sample. It is followed by predictive maintenance, triaging customer support requests, customer advocacy, and claims processing. Summaries of customer interactions and critical notes appear lower on the list, alongside sentiment analysis.

Overall, 40% of GenAI use cases cataloged by Databricks automate routine tasks related to the customer experience.

* On AWS (it’s in beta on Azure)