The Databricks Data + AI Summit 2026 is running right now — June 15 through 18 at the Moscone Center in San Francisco. With 30,000+ data and AI professionals in the room and tens of thousands more attending virtually from 150+ countries, this is where enterprise data strategy gets written.

The preview was useful. But previews are not announcements. Here is what actually shipped from Days 1 and 2.

Agent Bricks: The 99% Problem Gets a Platform

The biggest story out of DAIS 2026 is the expanded Agent Bricks platform. Databricks built Agent Bricks to address what they call the 99% problem: most teams focus on the core agent loop, but that is only 1% of what it actually takes to run agents in production. The other 99% is everything that breaks quietly — token capacity, deployment, security, evaluation, monitoring, context, and sharing.

The scale numbers make clear this is not a product in early innings. Since launch, over 100,000 agents have been built on Agent Bricks. The platform is now processing more than one quadrillion tokens per year from agent workloads. That is not a metric built on demos.

Enterprise customers AstraZeneca, 7-Eleven, Fox Corporation, and Block have shipped production agents on the platform. These are not companies doing pilots. They are running agents that do real work in regulated, high-stakes environments.

The expanded platform brings together model access, data context from the Lakehouse, identity-first governance through Unity Catalog, and automated evaluation that generates synthetic task data and LLM judges to optimize agents for cost and quality. The goal is a single platform from which a data team can build, deploy, evaluate, and govern agents without stitching together five different tools.

Catalog Federation: Governance Across the Cloud

The second major announcement was Catalog Federation moving from lab to general enterprise availability. This is governance work that matters enormously to data teams operating in multi-cloud environments or with legacy data infrastructure.

The concept: instead of forcing all data into Databricks, Unity Catalog becomes a single governance plane above external catalogs such as AWS Glue and Hive Metastore. You govern and query data managed elsewhere without copying it. Access control, lineage tracking, and audit logging apply consistently across catalog systems.

Mastercard demonstrated a production implementation, having successfully federated Databricks Unity Catalog to AWS Glue. That is a meaningful proof point for financial services and any organisation managing data under strict compliance requirements.

Unity AI Gateway: Governing Agents at Runtime

The Unity AI Gateway updates were the most operationally specific announcement of the summit. While governance of data assets is well-established in the Databricks ecosystem, governing AI agents at runtime is newer and harder.

The updated AI Gateway extends governance beyond data to the execution layer. It enforces access control, monitors usage, and audits activity across all MCP interactions in a workspace. That means the same policy engine covering your data now covers your agents and the tools they call.

Two additions stand out for enterprise teams: hard spend caps on external AI providers, and unified agent tracing. The spend caps are exactly what finance teams need — they stop requests when a budget is reached rather than surfacing an overage invoice at month end. The tracing layer captures model and MCP activity in a single governed telemetry layer, so you can actually see what your agents did, when, and at what cost.

Other Launches Worth Noting

Genie One is Databricks’ new agentic coworker for business teams. Rather than requiring SQL or Python knowledge, Genie One allows business users to interact with their data through natural language. It is positioned as the analyst layer for teams who should not need to open a notebook to answer a business question.

Lakebase reached general availability earlier in 2026 as Databricks’ serverless PostgreSQL product. It is designed specifically as the operational database layer for AI agents — the transactional data store that agents read from and write to as they work.

Lakehouse//RT delivers sub-100ms query latency at scale for real-time analytics use cases, closing a gap that previously required separate infrastructure.

What This Means for Business

The DAIS 2026 announcements collectively signal something that matters to business leaders who are not Databricks customers and never will be.

Enterprise AI has a governance problem. As AI agents proliferate inside companies, the question of who authorised what, how much was spent, what data was accessed, and whether the output can be audited is becoming urgent. The tools that solve this problem are being built now, and the vendors building them are winning enterprise deals.

For data teams, the direction is clear: agentic systems are production infrastructure, not experiment territory. The skill set is shifting from building data pipelines to building, evaluating, and governing agent pipelines. That shift is accelerating.

For business leaders who are not data experts, the implication is simpler: your data team is going to need new capabilities, and the platforms they work on are changing underneath them. The conversation about AI strategy is not separate from the conversation about data strategy.

Enterprise DNA exists at exactly this intersection — 220,000+ data professionals, structured learning paths from Power BI and Python through to AI-native workflows, and services that help businesses operationalise what their data teams learn. If your organisation is watching these announcements and wondering how to build toward them, start with Enterprise DNA’s business offerings.