Databricks Data + AI Summit 2025 just wrapped, bringing 22,000 people to Moscone and 65,000 more to the livestream — proof that data + AI is now a stadium sport. CEO Ali Ghodsi set the tone in his opening keynote, noting:
“Unified governance isn’t just a nice-to-have anymore — it’s more important than ever.”
That line echoed through every hallway conversation, breakout session, and partner demo. Below is our practitioner-friendly recap, tuned for the architects, analysts, governance owners, and execs who need to turn the DAIS hype into next quarter’s roadmap.
TL;DR: Three headlines that dominated the data and AI governance conversation #
-
Lakehouse → Lakebase → “Full-Stack” AI
Databricks’ new Lakebase (serverless Postgres on open storage) plus Apps and Agent Bricks re-frame the Lakehouse as a complete data-to-AI operating system.
-
Governance Moved from Slide Decks to Center Stage
Every keynote hammered home context, policy, lineage, and cost controls as table-stakes for trustworthy agents, exactly what Unity Catalog + partners like Atlan deliver.
-
Atlan is the Metadata Glue
We launched Data Quality Studio for Databricks, plus day-one integrations with Unity Catalog Metrics and Managed Iceberg.
Databricks’ macro narrative: A platform reimagined #
At the 2025 Databricks Summit, the big theme was completing the stack — transforming from a lakehouse into a full data-to-AI operating system.
Lakebase, a serverless Postgres engine built on open storage, brings operational and transactional data into the same lakehouse ecosystem. That means policies, lineage, and discovery now extend from dashboards all the way back to raw events. This shift dissolves traditional boundaries between OLAP and OLTP.
Agent Bricks, announced at Databricks Summit 2025, embeds cost and quality evaluation into agent development, Databricks introduced a built-in governance layer for enterprise AI. Instead of treating agents like toys, teams can now measure their output using baked-in “judge” models.
Databricks Apps let teams build full-stack apps right inside the Lakehouse, eliminating the need to push data out to yet another stack. Governance controls follow data natively, not through brittle API bridges.
Together, these updates align directly with Atlan’s vision of the “Active Metadata Lakehouse.” Governance is no longer bolted on, but the connective tissue.
Top 10 takeaways from Databricks Data + AI Summit 2025 that data leaders can act on #
1. Lakebase is now real #
Lakebase is Databricks’ reimagining of Postgres — serverless, open-format, and deeply integrated into the Lakehouse. It’s designed for workloads that used to live in app databases, and launch partners like Atlan, Retool, and Hex showed off early integrations.
2. Agent Bricks legitimizes AI agents for the enterprise #
Databricks baked “judges” into the platform — LLMs that measure agent quality and performance automatically. This shift moves agents from a hackathon novelty to something enterprise-ready.
3. Apps bring context to the forefront #
With Databricks Apps, developers can build interfaces that inherit Unity Catalog tags and permissions. One demo even showed glossary terms popping up inline in Chrome — giving business users direct access to governed definitions.
4. Unity catalog just got even more open #
Support for both Iceberg and Delta formats, two-way sync, and OSS contributions mean Unity Catalog is becoming the true backbone for open governance.
5. Atlan + Iceberg unlock real governance observability #
As a launch partner for Managed Iceberg, Atlan now plugs directly into Databricks to deliver metadata analytics, lineage visualizations, and AI policy enforcement — all with zero-copy data access.
6. Open formats go mainstream with Neon and Iceberg #
AI agents are spinning up more Postgres databases than humans, according to Neon. That means APIs, not GUIs, are the future of governance — and Atlan is ready.
7. Metadata is now the control plane #
Lineage graphics dominated every keynote at Data + AI Summit 2025. Governance isn’t just a compliance checkbox anymore — it’s how you scale agents safely. Ali Ghodsi said it best: “Unified governance is more important than ever.”
8. AI context just got real-time #
Databricks introduced GPU serverless infrastructure and upgraded vector search, enabling real-time context refresh for GenAI apps. It’s a major boost for tools like Atlan’s Data Quality Studio, which depends on up-to-date context.
9. Free edition + $100m training commitment #
Databricks launched a forever-free tier and pledged $100M to training. That’s a talent funnel every governance team needs to be ready for.
10. Customer stories drove the point home #
Mastercard talked about 600+ AI use cases powered by policy. GM showed how Atlan made their hybrid cloud architecture traceable end-to-end. Fox walked through their metadata plane. These weren’t sales pitches — they were blueprints.
Atlan moments that mattered 💙 #
Atlan didn’t just show up, we made context visible. We launched Data Quality Studio for Databricks, putting business-defined checks directly into Lakehouse pipelines. We partnered on Unity Catalog Metrics, transforming static KPIs into live, traceable assets. And we integrated with Managed Iceberg, powering observability for AI agents.
Our sessions with Fox and GM drew standing-room crowds, and our booth buzzed with deep questions on Iceberg, agent lineage, and policy propagation.
External Databricks Summit 2025 sessions to revisit #
While product announcements and roadmaps made headlines, some of the most resonant insights came from practitioners and industry leaders who showed what governance looks like in the wild. Here are four standout moments that drove home why context and trust are becoming table-stakes for enterprise AI:
Virgin Atlantic: Medallion architecture as a human-in-the-loop safety net #
Richard Masters from Virgin Atlantic shared how the airline’s governance framework goes beyond compliance. Their medallion architecture (Bronze, Silver, Gold layers) is now integrated with Unity Catalog’s row-level security and masking features — ensuring that only approved data makes it into production dashboards and AI pipelines. For safety-critical decisions like flight operations or maintenance predictions, this isn’t just data governance — it’s human-in-the-loop risk management.
Dario Amodei (Anthropic): Claude 4 and the AI alignment roadmap #
In a highly anticipated keynote, Anthropic’s CEO laid out how Claude 4 is pushing the boundaries of trustworthy AI. His focus? Alignment — not just with human values, but with enterprise governance constraints. From refusal behaviors to system-level guardrails, Dario emphasized that future AI agents will need to inherit metadata policies, not override them. For governance leaders, it was a call to prepare policy infrastructure now for the coming wave of virtual co-workers.
Joby Aviation: Democratizing real-time data with lakehouse lineage #
In a showcase of applied AI meets aerospace, Joby’s engineering team explained how their electric aircraft generate gigabytes of telemetry per minute — and how Databricks’ lakehouse platform makes that data accessible to hundreds of engineers and analysts. But what unlocked true speed? Lineage. With end-to-end tracking from raw sensor data to ML models and dashboards, Joby built a self-service platform where trust is embedded — not manually policed.
Matei Zaharia (Databricks): Unified governance with real benchmarks #
Databricks CTO and cofounder Matei Zaharia’s keynote was a technical deep dive into unified governance — but with numbers to back it up. He showed performance benchmarks for open formats like Delta and Iceberg under Unity Catalog, making the case that governance doesn’t have to be a tradeoff. Even more importantly, he introduced new lineage-aware cost controls and discovery tools — cementing that governance now extends beyond access into optimization and observability.
Take-action checklist for governance teams #
- Brief your execs: Lakebase + Agent Bricks = operational data is now governance territory.
- Audit metadata: Can your policies travel across Postgres, Delta, Iceberg?
- Try agent evaluations: Steal Databricks’ “judge” framework for your internal LLM pilots.
- Demo Atlan + Databricks: Book a deep dive into lineage, metrics, and glossary APIs.
- Get ahead of metrics lineage: Set up end-to-end tracing before your CFO asks where a number came from.
Wrapping up: Governance is the feature #
Databricks brought the house down with product launches — but what stood out most was a new attitude. Governance is more than a support function, but rather how AI earns trust.
FAQs: Databricks Data + AI Summit 2025 #
What is Lakebase and how does it extend the Databricks Lakehouse? #
Lakebase is a serverless Postgres engine that stores data on open-format cloud storage, letting teams mix operational and analytical workloads in one place. Because it lives inside the Lakehouse, policies, lineage, and discovery travel with every row instead of stopping at warehouse boundaries. This removes the usual divide between OLTP and OLAP, so dashboards and transactional apps share the same governance controls. For data leaders, it means fewer data copies and one security model to manage.
How do Agent Bricks make AI agents enterprise-ready? #
Agent Bricks introduces “judges,” small LLMs that automatically score the quality, cost, and reliability of every agent run. These metrics feed a continuous feedback loop so teams can tune prompts and guardrails without manual checks. Built-in evaluation shifts agents from experimental toys to production assets that meet governance standards. Organizations gain measurable trust signals that auditors and risk officers can track.
What new capabilities does Atlan bring after the Summit? #
Atlan launched Data Quality Studio for Databricks, embedding business rules and lineage views directly inside Lakehouse pipelines. It also became a day-one launch partner for Unity Catalog Metrics and Managed Iceberg, which lets teams observe metadata and policy flow across Postgres, Delta, and Iceberg tables. Two-way sync means any tag or lineage change in Databricks is reflected in Atlan and vice versa. The result is a single metadata plane that surfaces context in Chrome, SQL editors, and BI tools.
Why is Unity Catalog’s support for Iceberg and Delta so important? #
By adding full read-write support for both Iceberg and Delta formats, Unity Catalog turns into an open backbone rather than a format-locked catalog. Teams can choose the best table type for each workload while keeping consistent tags, policies, and ACLs. The two-way sync workflows are now open source, so organizations avoid vendor lock-in and keep governance portable. This openness also lowers the barrier for new tools that rely on shared metadata.
What should governance teams do first after reading the Data + AI Summit takeaways? #
Start by briefing executives that operational databases are now part of governance scope because of Lakebase. Next, audit whether existing policies can follow data across Postgres, Delta, and Iceberg, using Atlan or similar tools to map gaps. Pilot the judge framework from Agent Bricks on a low-risk internal LLM project to build evaluation muscle. Finally, schedule a joint Atlan-Databricks demo to explore lineage-aware metrics and glossary APIs in action.