Takeaways from the 2026 Databricks Virtual Partner Summit

The Wall Between OLTP and OLAP Is Finally Coming Down

I’ve spent about 20 years in data and during that time, one design rule has held: transactions live in one system, analytics live in another, and a pile of ETL sits between them. We built careers maintaining that pile.

After the Databricks Virtual Partner Summit, I believe that rule is starting to break. Two announcements are driving the change: Lakebase and Genie One. Here’s how I see them as a technologist, leaving the marketing aside.

Lakebase: Postgres that Lives in the Lakehouse

Lakebase is now generally available. It’s a fully managed, serverless PostgreSQL database built into the Databricks platform. There have been managed Postgres services for years, so that part alone isn’t new. What’s new is where the data lives.

Lakebase stores transactional data in open formats (Delta and Iceberg), governed by Unity Catalog. That’s the same storage layer the lakehouse uses for analytics. Databricks calls the pattern Lake Transactional/Analytical Processing (LTAP). In practice it means:

• One copy of the data: No CDC pipeline copying the operational database into the warehouse overnight, and no fights over reconciliation.
• One permission model: Unity Catalog governs the app’s tables and the analyst’s tables the same way.
• Git-style branching: You can create a full copy of a production database in under a second, then debug, test a migration, or let an agent experiment without touching live data. As of this month, snapshots and backup schedules can also be managed through APIs (beta), which matters for anyone building CI/CD around this.
• Lakebase Search (beta): Hybrid vector and full-text retrieval run natively in Postgres. You may not need a separate vector database next to your app database.

The part I care about most is agents. An agent that has to decide something needs two unique kinds of data: what’s happening right now (operational) and what usually happens (historical). When both sit in one governed store, you skip a whole class of stale data and access control problems. The new managed agent sessions and memory features on Lakebase point in the same direction.

Genie One: An Agent that Looks Things Up Instead of Guessing

Many enterprise AI tools sound confident and get the numbers wrong. Genie One takes a different approach. It answers by running SQL against governed data, and it draws business context from Genie Ontology.

The Ontology is a live layer that learns metric definitions, business terms, and relationships from your tables, queries, dashboards, pipelines, and connected apps.

A line from Databricks’ CEO, Ali Ghodsi at the ’26 DAIS summit, stuck with me: “If you’re a CFO and AI can’t tell you why margins changed, that’s not an AI problem, that’s a context problem.”

That matches what I see with clients. The model is rarely what holds a project back. Missing semantics are the culprit.

Here’s what caught my attention technically:

• Grounded answers across 50+ workplace apps, including Slack, Teams, Jira, Confluence, SharePoint and Google Drive, with Unity Catalog permissions enforced.
• Output beyond chat. It produces documents, reports, scheduled tasks, and alerts. Genie Agents can turn a good conversation into a reusable agent without code.
• The Genie One Model Context Protocol (MCP) server went generally available (GA) on September 25, 2026. Other agents and tools can now use Genie One as a governed source of truth through a standard protocol. For people designing agent systems, I think this is the most important update of the month.
• No per-seat pricing, with a monthly free usage allowance per user. That lowers the barrier to broad adoption.

Also on the roadmap: Genie Code (GA and now free) for data engineering work, and Genie ZeroOps (private preview), a background agent that detects pipeline failures and proposes fixes for a human to approve.

Why These Two Belong Together

Look at the two launches side by side. Lakebase removes the gap between operational and analytical data, and Genie One removes the gap between that data and the people who make decisions. Governance runs through both: one catalog, one set of permissions, one ontology.

That’s the architecture I’d want for agentic workloads. Agents run on data that’s current, governed, and tied to the same meaning your finance team uses.

What I’m telling clients

1. Take stock of your OLTP-to-lakehouse pipelines. Each one is a candidate to simplify with Lakebase.
2. Invest in semantics now. Unity Catalog metrics, a business glossary, and clean ownership are what make Genie One accurate. Without them, it’s just a chat window.
3. Start with a small, high-value domain. Pick one question the business asks every week and see whether Genie One answers it correctly.
4. Design for MCP. Your future agents will call Genie One as a tool, so plan for that from the start.

For decades we accepted separate systems as a law of physics. It was never a law. It was a trade-off we lived with, and I don’t think we have to anymore.

Which pipeline would you retire first? Connect with our Databricks experts to transform your data estate.