Data + AI Summit 2026: Day 1: The 12 Announcements That Redefine the Databricks Platform

The modern data architecture has always involved an uncomfortable trade-off: to get millisecond latency, you had to copy your data to a separate serving layer, a proprietary copy isolated from your governance, your lineage, and your analytics stack. Two systems, two pipelines, two costs, two risks.

Lakehouse//RT eliminates that.

Powered by Reyden, a brand-new execution engine built from scratch for high-concurrency, low-latency workloads, Lakehouse//RT delivers millisecond performance directly on your lakehouse data, no data movement, no copying, no second layer.

The compute model is new too: AUTO sizing (no t-shirt sizes) and incremental autoscaling node by node, not doubling whole instances. Preview customers including Meta, SES, Cisco, Bally’s, Enverus, and PointClickCare saw improvements ranging from 4× to 100× on real production workloads.

Unity Catalog governance applies automatically, no need to redefine policies in a second system because there is no second system.

Lakeflow was already Databricks’ data engineering stack, declarative pipelines, orchestration, ingestion. With Lakeflow Agentic Data Engineering, the platform takes the leap toward pipelines that build, monitor, and repair themselves using agents.

The result is data engineering that scales team productivity without growing headcount proportionally, engineers stop doing plumbing and start focusing on business logic.

Genie is Databricks’ natural language interface for analytics, any user asks questions in plain text and gets answers about their data. The problem until today: configuring and maintaining a Genie Space required continuous work from the data team.

Genie ZeroOps eliminates that operational friction. The system automatically manages semantic context, updates business definitions as data evolves, and scales availability without anyone needing to intervene manually. Data teams go from maintaining Genie to simply using it.

LakeBase is Databricks’ managed Postgres, a transactional and operational database unified with the lakehouse. LakeBase Search adds semantic and vector search capabilities directly on top of LakeBase, designed specifically for AI agents to retrieve relevant information without additional infrastructure.

Traditional CDPs have the same problem as serving layers: they copy data, isolate context, and create technical debt. CustomerLake is Databricks’ answer: an agentic customer data platform built directly on the lakehouse, unifying user profiles, behaviors, and real-time signals without data movement.

Databricks unveiled three interconnected capabilities that transform Genie into far more than a data chatbot:

Runaway generative AI spend is one of the biggest operational problems of 2026. Teams deploy dozens of models, agents, and MCP servers with no central visibility or control. Unity AI Gateway is the answer: a centralized control plane that governs all AI activity at runtime.

Unity Catalog receives a set of updates focused on supporting the agentic world: cataloging of agent tools, extended lineage for AI decision traces, access policies for MCP servers, and improvements in managing non-human identities (service principals and autonomous agents).

The headline announcement from Kasey Uhlenhuth on the DAIS 2026 stage: Agent Bricks is Databricks’ framework that organizes everything you need to move AI agents from pilot to production around three pillars.

Delta Sharing was Databricks’ bet on open data exchange between organizations. Open Sharing is the next evolution: a protocol for sharing not just data, but also tools, models, and agent context across organizations, the connective tissue of the cross-enterprise agentic ecosystem.

SecureConnect closes the security piece: it allows the exchange to happen over private networks, without data crossing the corporate perimeter, eliminating the risk of accidental exposure while maintaining the agility of collaboration.

The final announcement of the Keynote and one of the most disruptive for business teams: Genie App Builder is Databricks’ first native enterprise vibe-coding tool. Any business user describes the app they need in natural language (an intake form, a weekly dashboard, a scenario calculator) and gets it deployed, governed, and connected to their data without touching code.