Databricks launches AI coding benchmark and expands enterprise AI platform with CustomerLake

Databricks released an internal benchmark that evaluates AI coding agents on a realistic codebase of millions of lines of Python, Go and TypeScript. The results show open‑source models matching top proprietary options while highlighting that per‑token pricing can be misleading for overall cost, informing the company's ongoing integration of AI coding tools.

At the Data + AI Summit 2026, Databricks announced a shift toward enterprise‑scale agentic AI, unveiling an Agentic Customer Data Platform called CustomerLake. The platform embeds the four imperatives of Choice, Context, Cost and Control, integrates AI governance directly into execution, and adds security information and event management capabilities. New components such as the Omnigent meta‑agent layer aim to orchestrate and govern multiple AI agents at scale.