Azure Databricks took a decisive leap toward real‑time analytics and frictionless data sharing at the Data + AI Summit 2026 in San Francisco, where Databricks laid out a vision that binds the lakehouse tighter to the Microsoft ecosystem. The four‑day event, held June 15–18, introduced Lakebase for unified governance, Lakehouse//RT for low‑latency workloads, native OneLake integration, a Genie natural‑language assistant for Microsoft 365, and write‑back support directly into Excel spreadsheets. Each addition points to a strategy that no longer treats the data platform as a standalone island but as an embedded control plane for enterprise AI.

Microsoft shops running Power BI, Excel, and Fabric suddenly had several reasons to lean harder into Azure Databricks. The updates arrive as corporate IT departments grapple with agentic AI — systems that not only generate answers but can take action across databases, APIs, and business applications. Governance, Databricks executives argued on stage, must come first, and the new features reflect that priority by centering access controls, auditing, and lineage inside the lakehouse itself.

Lakebase: one governance layer to rule them all

The centerpiece of the governance overhaul is Lakebase, a metadata and policy engine that unifies catalogs across Databricks workspaces, Microsoft Fabric, and external data sources. Early adopters told attendees that Lakebase eliminates the patchwork of Unity Catalog configurations they previously had to stitch together manually. Administrators now define row‑level filters, column masks, and data‑sharing rules once, and Lakebase propagates them across any connected system — including OneLake, SQL Server, and third‑party data lakes.

Baked into Lakebase is a new “agency passport” for agentic AI workloads. Every AI agent that queries a dataset inherits the same identity and permission constraints as the human user who launched it. If a financial analyst triggers an agent to fetch sales projections, the agent cannot inadvertently see HR data. Databricks demonstrated a scenario where an agent built with Mosaic AI autonomously created a demand forecast, posted it to a Teams channel via a Custom connector, and wrote the final numbers back into an Excel spreadsheet — all while Lakebase enforced field‑level masking on individual customer rows.

Lakehouse//RT: real‑time finally arrives without a separate stack

For years, real‑time analytics on Databricks required a side‑car serving layer or a completely separate stream processor. Lakehouse//RT removes that friction by embedding a low‑latency query engine directly into the lakehouse. The system captures change data from transactional databases, processes it through Structured Streaming, and makes fresh rows queryable in under 250 milliseconds, according to benchmarks shared on the expo floor. Microsoft sources confirmed that the engine is co‑engineered with Azure Synapse to run on the same compute substrate, letting customers fail over seamlessly between Databricks and Fabric real‑time nodes.

A demo that drew applause showed a logistics company running continuous ingestion of IoT telemetry from 15,000 delivery vehicles. Lakehouse//RT surfaced route deviations within half a second, triggered automated alerts in Power Automate, and logged every decision for compliance — all inside a single notebook. That level of responsiveness, now available as a GA feature, could eliminate the need for dedicated streaming databases in many Azure architectures.

OneLake goes bidirectional

Databricks and Microsoft had already established short‑cut access to OneLake from Databricks workspaces. The 2026 summit made the relationship symmetric: OneLake now recognizes Databricks‑managed tables as native delta‑parquet assets, meaning a Fabric Data Factory pipeline can directly transform a table that Databricks govern through Lakebase. IT teams no longer need to export snapshots or maintain duplicate copies; the lakehouse and Fabric share the same data, metadata, and security model.

This tight coupling extends to Power BI as well. Databricks showcased a “Direct Lake” connection where Power BI reports pointed at live, streaming‑enriched datasets inside Lakehouse//RT, rendering visuals that refreshed every second without any import or cache refresh. The combination of Lakebase governance and Direct Lake performance gives organizations governed self‑service analytics that still meet regulatory audit standards.

Genie for Microsoft 365: natural language meets the productivity stack

Databricks first previewed Genie, its natural‑language interface, in 2025. At this year’s summit, the company embedded Genie directly into Microsoft 365 applications. Users can now type a plain‑English question in an Outlook add‑in panel — “What was the win rate for the EMEA region last quarter?” — and Genie queries governed lakehouse datasets, returning a formatted answer along with a provenance trail that links every figure back to a specific column and row.

Early testers noted that Genie for Microsoft 365 respects the user’s existing Microsoft Graph identity, so it already understands organizational context and access boundaries. A sales manager typing in a Teams chat can invoke Genie with an @Databricks mention, receive a card containing the answer, and pin it to a shared channel tab. The integration pulls heavily on the new Lakebase agency passport, ensuring that conversational AI never over‑shares sensitive metrics.

Excel remains the unrivalled front end for business users, and Databricks finally gave them a direct write‑back path. From within a standard Excel workbook, a user connected to a governed lakehouse table can edit a cell and push the change back to the source — complete with validation rules defined in Lakebase. The feature works with both Excel desktop and Excel for the web, using the same OAuth token that Azure Active Directory assigns.

On‑stage demos showed a budget administrator adjusting a marketing spend figure of $1.2 million to $1.5 million. A few seconds later, the updated value appeared in a Power BI dashboard and triggered a Slack notification (via a Custom agent) that the finance team needed to re‑sign a purchase order. Databricks called the workflow “bidirectional governance,” meaning the lakehouse treats the spreadsheet as a first‑class client rather than an export‑only afterthought. For enterprises that have struggled with “Excel‑sprawl,” the write‑back capability could be a governance knockout punch.

Custom connectors and agentic AI pipelines

The “Custom” piece teased in the summit rundown refers to a new framework that lets developers build reusable connectors for any REST API, GraphQL endpoint, or legacy system. These connectors become first‑class lakehouse objects, inheriting Lakebase policies, and can be dragged into agentic AI workflows built with Databricks’ Mosaic AI. A Custom connector for Salesforce, for example, can pull opportunity records, join them with lakehouse product‑usage data, and pass the merged set to an agent that scores accounts — all without writing a single line of glue code.

Databricks’ CTO demonstrated a fleet of agentic AIs that orchestrated a full quarter‑end financial close: one agent queried SAP for actuals, another pulled forecast revisions from an Excel write‑back table, a third ran anomaly detection inside Lakehouse//RT, and a fourth drafted an exec summary that appeared in a Word document inside Microsoft Teams. The entire run was audited by Lakebase and surfaced in a new “Agent Activity Monitor” that shows, for every AI action, the initiating user, the dataset touched, and the policy applied.

Governance first, then innovation

The summit’s unofficial slogan — “Governance first for agentic AI” — wasn’t just marketing. Every feature that Databricks launched ties back to one of three governance pillars: identity‑driven access (Lakebase agency passport), data‑lineage tracking (the provenance trails in Genie and Excel write‑back), and policy‑enforced quality control (validation rules that propagate from lakehouse to spreadsheet and back). In a breakout session, a healthcare CIO shared that the unified governance had reduced their time to onboard a new AI use case from six weeks to four days, simply because security and compliance checks were already embedded.

For Windows enthusiasts who manage enterprise environments, the message is clear: Azure Databricks is positioning itself as the command center for safe agentic automation. The deep Microsoft integration means that tools users already interact with daily — Excel, Outlook, Teams, Power BI — become the interactive endpoints for governed, auditable AI. Windows administrators can manage the entire stack through Azure Active Directory conditional access policies, and every data touchpoint logs into a single Azure Monitor workspace.

What this means for the Windows enterprise

Windows 11 already ships with native Fabric and Azure Data Studio integrations, and the Databricks updates slot neatly into that roadmap. The Excel write‑back feature, in particular, could transform how finance, supply chain, and marketing departments interact with central data warehouses. Instead of emailing spreadsheet attachments that quickly diverge from the source, teams will edit governed cells that stay in sync with the authoritative lakehouse. Genie for Microsoft 365 lowers the barrier to analytics without sacrificing governance, making it plausible that line‑of‑business users can ask complex questions without queuing up a ticket for the data engineering team.

IT professionals should start planning for the operational impact. The bidirectional OneLake integration may reduce storage costs by removing redundant copies, but it will require mapping existing Unity Catalog hierarchies into the new Lakebase model. The real‑time engine inside Lakehouse//RT is co‑engineered with Synapse, so organizations already on Fabric real‑time intelligence can mirror workloads and gradually shift to the Databricks console. Early adopters advised that the migration path is not a lift‑and‑shift; it demands rethinking how streaming jobs are partitioned and monitored.

Agentic AI pipelines running on Custom connectors will also force a review of API governance. Because connectors inherit lakehouse policies, a misconfigured policy could inadvertently expose data to an agent. Databricks released a set of pre‑flight validation rules that simulate an agent’s blast radius before it’s promoted to production, and Microsoft Defender for Cloud will include specific recommendations for Azure Databricks agent workloads by Q4 2026.

Looking ahead

Databricks didn’t specify pricing for the new features beyond confirming that Lakebase is included in the existing Unity Catalog license, while Lakehouse//RT and Excel write‑back will follow a consumption‑based model. The Genie for Microsoft 365 add‑in is free for users with an active Databricks SQL warehouse, though advanced natural‑language model inference may consume GPU‑attached pricing units.

The company’s roadmap slide hinted at “Lakehouse//Edge” for on‑premises Windows Server nodes, potentially enabling the same real‑time engine to run in a factory-floor Windows IoT deployment. While details were vague, it signals that Databricks sees the entire Windows estate — from handheld scanners to Azure VMs — as fair game for governed, agentic analytics.

For Windows‑focused IT journalists, the summit delivered a single, resonant throughline: data governance is no longer a pre‑requisite checklist; it’s the product. By embedding governance into the tools people already use, Databricks and Microsoft are betting that the next wave of enterprise AI will be defined by trust, not just speed. The enhancements announced in San Francisco suggest that bet is already paying out.