Last month, San Francisco wrapped the busiest Databricks Data + AI Summit yet with over 31,000 attendees, 400 hours of session content, and two massive keynotes. But the volume of announcements wasn’t the story. The story was the through-line connecting nearly all of them: agents, plus governance, plus real-time data, unified under Unity Catalog.
Databricks framed the shift in a single sentence: the catalog is moving from a “system of record” to a “runtime decision-maker for AI.” At Lovelytics, that framing matches exactly where we’ve been steering our customers, where governed agentic AI is no longer a future opportunity but something every organization can – and should – start building today. Here’s our read on what mattered most, and what it means for your team.
Genie grows up: from chat assistant to AI coworker
Genie evolved from a conversational analytics assistant into what Databricks now calls a “data-smart AI coworker.” Genie One connects across the entire data estate — native connectors, Lakehouse federation, Lakeflow Connect, and two-way integrations with the everyday tools where work actually happens (Gmail, Slack, Teams). It adds true agentic-cowork capabilities: scheduling, alerts, monitoring, document creation, custom skills, and custom MCP support, with native @mention access inside Slack and Teams plus new iOS and Android apps.
Two companions round it out. Genie Agents turn Genie Spaces (1M+ created) into curated, domain-specific agents that take autonomous multi-step action and reason over unstructured documents as well as tables. Genie Ontology is an automatic, living context graph that extracts meaning from tables, queries, dashboards, and pipelines, using a PageRank-like approach to weigh source authority and freshness — while enforcing source-native permissions automatically. The benchmark Databricks shared is hard to ignore: on its internal 28-question enterprise suite, Genie answered 84.5% of questions correctly on the first attempt versus 52.4% for the strongest general-purpose coding agent, at roughly 2x the speed.
Lovelytics’ take: the benchmark is impressive, but the unsung hero is Genie Ontology. We’ve seen firsthand that AI assistants succeed or fail on the quality of the business context behind them. Databricks has become the one-stop-shop for data for many organizations, and now turning that enterprise data into AI context is seamless.The teams that win with AI won’t be the ones who simply turn it on, they’ll be the ones who’ve invested in clean semantics, a well-designed medallion architecture, and curated domains for the ontology to learn from.
Agent Bricks becomes a full agent platform
One year after launch, Agent Bricks has grown to 100k+ agents built and over a quadrillion tokens processed per year (AstraZeneca, 7-Eleven, Fox Corporation, and Block among them). Databricks reframed it from a builder into a complete platform, with a memorable argument: the core agent loop is only 1% of the work, while the other 99% is the hidden technical debt of token capacity, deployment, security, evaluation, monitoring, context, and sharing.
The platform now centers on choice, context, and control. On choice, every frontier and open-source model lives in one place (OpenAI, Anthropic, Gemini, Qwen, the newly added Kimi, and Grok via a new SpaceX partnership), with support for LangGraph, Agno, CrewAI, the Claude Code SDK, and OpenAI Agent SDKs. On context, MCP support inside Unity Catalog lets agents securely reach Google Drive, Jira, Slack, and GitHub, backed by an agent memory service on Lakebase and Document Intelligence in SQL. And on control, the new Unity AI Gateway governs agents, models, MCPs, and skills with fine-grained access, per-user budgets, and intelligent routing.
Lovelytics’ take: “99% of the work is everything around the agent loop” is the most honest thing Databricks said all week. It’s also why so many promising agent pilots never reach production. We help customers treat the Unity AI Gateway as the control plane it’s meant to be — standing up governed model choice, MCP connectivity, auditability, and spend controls before the first agent ships, not after.
Unity Catalog: governing what agents do, not just what they access
More than 14,000 organizations now govern data and AI on Unity Catalog, and the 2026 updates push governance into the agent runtime itself. Contextual Service Policies can allow, deny, or require approval for specific actions — writing to a sensitive folder, pushing code — with built-in guardrails for PII, prompt injection, and unsafe content. A new Governance Hub gives teams a command center to monitor posture and prioritize remediation across data, AI, cost, and performance, and an open ecosystem of integrations is arriving with CrowdStrike, Palo Alto Networks, Zscaler, Okta, and others.
On the context side, Glossary and Domains establish shared business meaning, governed Metrics deliver reusable KPI definitions (with Power BI and Tableau import), and external lineage now spans non-Databricks systems. On choice, a new four-level namespace gives every asset a single cross-cloud address under one set of policies and one audit trail.
Lovelytics’ take: governance has quietly become the hardest — and most valuable — problem in AI. We wrote recently that the rules are changing fast, and DAIS 2026 confirmed it: the question is no longer just “can this user see this table?” but “what is this agent allowed to do, on whose behalf, and at what cost?” Designing those policies deliberately, up front, is foundational work we believe every enterprise should be doing today.
CustomerLake: the agentic CDP — and a launch we’re proud to be part of
The launch closest to home for us is CustomerLake, a new agentic Customer Data Platform embedded natively in the Databricks lakehouse — and Databricks’ formal entry into the marketing industry. It brings Customer 360, identity resolution, audience building, activation, and personalization to the governed data and AI foundation enterprises already use, with no copying or duplicating of sensitive data. It’s now in Private Preview, built on three principles: Embedded, Democratized, and Autonomous. Lovelytics is proud to be part of the CustomerLake launch, and we believe it marks what our own Murray Williams has called “the third age of the CDP.”
Two new agent types anchor the platform. Profile Agents turn raw customer data into business-ready Customer 360 profiles, powered by Agentic Identity Resolution (AIR) that blends deterministic, probabilistic, and agentic workflows. Campaign Agents move teams beyond static, one-off campaigns toward continuous, agent-driven “infinity campaigns” — building audiences, recommending next-best actions, activating across channels, and optimizing around business goals, while humans set the strategy and guardrails. It’s interoperable through Unity Catalog governance and Lakehouse Federation (Snowflake, BigQuery, and more), with a launch ecosystem that includes Adobe, Meta, Braze, Acxiom (whose integration we’re helping build), Epsilon, The Trade Desk, and LiveRamp — and services partners including Lovelytics. Early customers already include HP, Getnet by Santander, and Zé Delivery (AB InBev).
Lovelytics’ take: our view is that good architectural planning is the cornerstone of any Customer 360 initiative — a gold-layer single view of the customer, a clear identity strategy, privacy and consent built into the foundation, and a carefully scoped real-time plan. Traditional CDPs often became silos, while so-called “composable” CDPs required constant management and data movement. As a launch partner, we’re mapping these proven methodologies directly to CustomerLake’s new Profile Agents and Agentic Identity Resolution, so customers are structurally ready to let these agents drive their identity strategy forward. You can read our full perspective in Databricks CustomerLake Ushers in the Third Age of the CDP.
Real-time, unified: Lakehouse//RT and LTAP
Two of the most consequential infrastructure launches close the gap between operational and analytical data. Lakehouse//RT (Beta) brings real-time analytics directly onto the governed lakehouse powered by a new compute engine, Reyden, built for the concurrency and latency that agentic workloads demand. It delivers millisecond query latency at tens of thousands of concurrent users and agents (as low as 10ms on smaller datasets, sub-100ms at 12,000 queries per second), with customers reporting up to 16x better performance than existing real-time stacks. Every query runs natively within Unity Catalog with no separate permissions layer, no proprietary formats, and no sync/CDC pipelines.
Alongside it, LTAP (Lake Transactional/Analytical Processing) unifies OLTP and OLAP on a single copy of data, eliminating ETL, replicas, and hidden pipelines by design. Databricks calls it the world’s first LTAP platform, pairing Lakebase (serverless Postgres on open object storage) with the Lakehouse under one governance model. Lakebase now serves thousands of customers and handles 12 million database launches per day, and is adding cross-region disaster recovery, git-style branching against production data, and autonomous database operations.
Lovelytics’ take: the “two databases for one job” pattern has quietly taxed enterprises for years with duplicated data, brittle pipelines, and stale analytics. Even as we’ve started adopting Lakebase more and more as the foundation for AI-native applications on Databricks Apps we’ve seen the burden of managing both first hand. Collapsing transactions and analytics onto one governed copy is the kind of simplification that pays dividends well beyond performance, and it’s a natural fit for the real-time, event-driven customer experiences we help our clients design.
The security lakehouse: Databricks to acquire Panther
Databricks announced its intent to acquire Panther, a leading AI SOC platform, to advance its “security lakehouse” vision — a new category aimed squarely at displacing the legacy SIEM with an agentic approach. It’s Databricks’ third security acquisition (after Antimatter and SiftD.ai) and is already trusted by security teams including Anthropic. The argument is familiar: legacy SIEMs are held back by high cost, limited data, and manual workflows, so most organizations analyze only a fraction of their security data.
Panther replaces closed SIEM stacks with agentic SOC workflows that auto-triage alerts and propose next steps, more than 100 pre-built, deeply parsed integrations across cloud, identity, endpoints, networks, and SaaS, and detection-as-code (from the creator of the open-source StreamAlert project). It plugs directly into Lakewatch, Databricks’ lakehouse-native agentic SIEM, so defenders can investigate every alert and fight AI-driven attacks with AI.
Lovelytics’ take: security and governance are converging fast. As agents gain the ability to act, the same Unity Catalog policies that govern data access and the Unity AI Gateway that governs agent behavior become the natural foundation for detection and response. We help customers treat security telemetry as another governed lakehouse workload rather than a bolt-on, so investigations draw on the full context of the estate.
Apps on Databricks Marketplace
Now in Public Preview, Apps on Databricks Marketplace lets customers discover, install, and run third-party data and AI applications directly inside their secure workspaces — “the application comes to your data” instead of moving data to the vendor. It launched with 20 partners. A workspace admin reviews the requested permissions and resource bindings, then one-click installs onto serverless compute with a dedicated URL.
Crucially, it’s governed and IP-protected: apps run in an isolated sandbox inside the consumer’s account, inheriting Unity Catalog governance, with egress controlled via the Serverless Egress Gateway and provider apps shipped as closed-source containers. Native ties to SQL Warehouse, Lakebase, Model Serving, Genie, and Agent Bricks make many apps AI-ready out of the box, while providers can publish once and reach thousands of customers with no per-customer infrastructure.
Lovelytics’ take: “bring the app to the data” is the logical endpoint of the zero-copy thesis that also underpins CustomerLake. For our customers, it means evaluating new capabilities without standing up new environments or exporting sensitive data — and it raises the bar on governance, since every installed app inherits the Unity Catalog policies you’ve already designed.
Lakeflow: a new era of agentic data engineering
Lakeflow is Databricks’ unified platform for all of data engineering — ingestion, transformation, and orchestration — fully governed by Unity Catalog, giving AI agents a single source of trusted, real-time context to both build and operate pipelines. Genie Code is now integrated across Lakeflow, and Lakeflow Designer (GA) adds a visual, no-code, AI-powered drag-and-drop canvas where every flow runs natively on a production Spark Declarative Pipeline with zero translation loss. Genie ZeroOps runs in the background, performing root-cause analysis from quality metrics, error logs, and lineage and proposing fixes validated in a governed sandbox.
The connectivity story is just as deep: Lakeflow Connect now offers 100+ native managed connectors (Jira, GitHub, Confluence, SharePoint, Google Drive, HubSpot, the major ad platforms, and more) plus a Free Tier, while Zerobus Ingest enables Kafka-free ingestion of high-volume event data at sub-5-second latency and up to 100MB/s. Real-Time Mode for Spark Declarative Pipelines reaches end-to-end latencies as low as 5ms without a second engine. (Panasonic cited 50% faster Power BI refresh; Meta uses Zerobus to bridge on-prem to cloud.)
Lovelytics’ take: the pattern across every one of these launches is the same — collapse the integration tax and let teams build on a single governed copy of the data. But agent-assisted pipelines are only as trustworthy as the lineage and semantics beneath them, which is exactly the foundational work we prioritize so that Genie Code and Lakeflow Designer produce pipelines you can actually stand behind.
Also worth your attention
- Genie Code — a specialized agent for data and ML engineering with a full-page command center (one customer runs 15+ parallel threads daily), native ML-stack integration, and the forthcoming Genie ZeroOps for autonomous production operations.
- AI Platform — AI Runtime (Public Preview) offers 2–3-click access to serverless A10 and H100 GPUs, and High-QPS Model Serving reaches 300K+ QPS at under 10ms p99 latency; customers like Grammarly and GoGuardian have cut serving costs 90%+.
- Security & compliance — Automatic Identity Management is GA on AWS and GCP, Context-Based Ingress brings zero-trust access, and compliance expands across HITRUST, ISMAP, and FedRAMP High on Azure.
The bottom line
Databricks is pushing hard into governed, agentic AI: the catalog becomes a runtime control plane, and the platform aims to be the single place to build, run, and govern agents and real-time workloads. Unity AI Gateway and Agent Bricks are the most strategically relevant launches for teams building governed agents, while Genie One, Genie Code, and Genie Ontology lower the barrier to putting that data to work.
For Lovelytics, this only strengthens our core belief: that agents are only as good as the architecture, semantics, governance, and data beneath them. That foundational heavy lifting is what makes the lakehouse genuinely work for the business, and it’s what we do every day for our customers.
Want help turning these announcements into an actionable plan?
Whether you’re standing up governed agents on the Unity AI Gateway or accelerating your move to an AI-native CDP with CustomerLake, reach out to the Lovelytics team — an eight-time Databricks Partner of the Year and a CustomerLake launch partner — to get started.
