Databricks built Agent Bricks to take agents out of the lab and into production, and at the 2026 Data + AI Summit the company showed how far that has gone: more than 100,000 agents built and over a quadrillion tokens processed every year. The hard part, Databricks found, was never the core agent loop. It was the surrounding 99%: governing the data agents touch, grounding them in real business meaning, evaluating quality, and controlling cost. That meaning rarely lives in one place. Most enterprises run Databricks alongside Snowflake, dbt, Tableau, Salesforce, SAP, and more, which is exactly where an enterprise context layer earns its keep.
Quick facts
Permalink to “Quick facts”| Attribute | Detail |
|---|---|
| What it is | Databricks’ platform to build, evaluate, and optimize AI agents on governed data |
| Announced | Expanded developer agent platform, June 16, 2026 at Data + AI Summit |
| Category | Enterprise agent platform on the lakehouse |
| Who it’s for | Data and AI engineers, platform teams, and developers building production agents |
| Key benefit | One governed surface for the 99% of agent work beyond the core loop |
| Works with | Unity Catalog, Genie Ontology, MCP, LangGraph, CrewAI, Agno, Claude Code SDK |
| Reported scale | 100K+ agents built; 1+ quadrillion tokens processed per year |
| How Atlan complements it | Supplies governed, cross-estate context to Agent Bricks agents via MCP and the Enterprise Data Graph |
Inside the expanded Agent Bricks platform
Permalink to “Inside the expanded Agent Bricks platform”The core agent loop, Databricks told the Summit, is only about 1% of the work. The other 99% is the hidden technical debt of agentic systems: token capacity, deployment, security, evaluation, monitoring, context, and sharing. Agent Bricks is Databricks’ answer to that 99%, organized around three pillars the company repeated throughout the event: Choice, Context, and Control.
On Choice, Agent Bricks supports a wide range of frontier models (OpenAI, Anthropic, Gemini, Qwen, and Kimi, plus Grok via a SpaceX partnership) and custom models tuned through prompt optimization, fine-tuning, and reinforcement learning. On harnesses, it runs LangGraph, Agno, CrewAI, the Claude Code SDK, and OpenAI Agent SDKs, alongside Databricks’ open-source Omnigent meta-harness. On Context and Control, it leans on Unity Catalog for governance and Genie Ontology for business meaning.
Build, evaluate, and optimize
Permalink to “Build, evaluate, and optimize”Agent Bricks is built around a lifecycle, not a single feature. Teams build agents with their chosen model and harness, evaluate output quality with Agent Evaluation (AI-assisted assessments plus a UI for human feedback), and optimize quality and cost using synthetic data, custom evaluation, and automated tuning. The result is a path from prototype to a production agent that is measured, not assumed.
| Capability | What Agent Bricks provides |
|---|---|
| Build | Model choice, agent harness support, managed agent memory via Lakebase |
| Evaluate | Agent Evaluation with AI-assisted scoring and human-feedback UI |
| Optimize | Synthetic data, custom eval suites, automated quality and cost tuning |
| Connect | Native MCP access to APIs, databases, and SaaS (Google Drive, Jira, Slack, GitHub) |
| Govern | Unity Catalog registry for agents, tools, and models; Unity AI Gateway runtime control |
Is your data estate ready for agents like these?
Agents built in Agent Bricks reason over the context you give them. Check how ready your estate is to supply governed, cross-system context.
Assess Your ReadinessHow Agent Bricks relates to Genie and Unity Catalog
Permalink to “How Agent Bricks relates to Genie and Unity Catalog”Agent Bricks does not work alone. It sits inside a stack where Unity Catalog governs the agents, tools, and models, and Genie supplies the conversational surfaces and the business meaning underneath them. Understanding how the pieces fit makes the role of each clearer.
Genie Ontology is Databricks’ context layer for Genie: a living web of an organization’s knowledge from data, docs, tags, apps, and people, embedded directly into retrieval and planning. Databricks reported that this context delivered materially higher accuracy than standard retrieval on multi-step work. Genie One, the agentic coworker announced alongside the expanded platform, is one of the surfaces that draws on this foundation.
| Component | Role in the Databricks agent stack |
|---|---|
| Agent Bricks | The platform to build, evaluate, and optimize agents |
| Genie Ontology | The business context layer that grounds Genie agents |
| Genie One | The agentic coworker surface for business teams |
| Unity Catalog | Governance for agents, tools, models, and the data they touch |
| Unity AI Gateway | Runtime governance: spend caps, routing, and guardrails |
Genie Ontology and Unity Catalog are excellent inside Databricks. The open question for most enterprises is what happens to the context that lives everywhere else, and how it reaches the agents you build in Agent Bricks.
Why context is the real constraint
Permalink to “Why context is the real constraint”According to Atlan AI Labs research, 83% of AI pilots never reach production, and the gap is context, not the model. An agent that knows the lakehouse schema but not the company’s fiscal calendar, or that resolves “revenue” differently than the finance team does, will be confidently wrong. This is why AI agents need an enterprise context layer: the model is rarely the bottleneck, the meaning is.
See the Context Layer in action
Watch how teams deliver governed, cross-estate context to the agents they build on Databricks and beyond.
Watch Context Layer LiveHow Atlan’s context layer extends Agent Bricks across the estate
Permalink to “How Atlan’s context layer extends Agent Bricks across the estate”Atlan is the context layer for AI: the governed infrastructure that delivers enterprise knowledge to every model, every agent, and every team from a single source of truth. It layers on top of your existing stack, so most enterprises run Atlan alongside Databricks Unity Catalog rather than rebuilding context from scratch. The frame is better together: Databricks brings the data and the horsepower, Atlan brings the meaning, governed and live, across the entire data and AI ecosystem.
Genie Ontology grounds agents in Databricks context. Atlan extends that grounding to the whole estate. A Genie or Agent Bricks agent asked about “customer churn” often needs context from Salesforce for CRM definitions, dbt for transformation logic, and Tableau for metric definitions, not just the lakehouse. Atlan unifies context across all of those systems and serves it back to Databricks agents, so the context an agent reasons over is as wide as the business itself.
The four products that supply cross-estate context
Permalink to “The four products that supply cross-estate context”Enterprise Data Graph: 80+ connectors and column-level lineage build a living graph of assets and relationships across the whole estate. Agent Bricks reaches across the systems you connect to Databricks; the Enterprise Data Graph reaches across every system, including the ones outside it.
Context Agents: AI teammates that auto-generate descriptions, link glossary terms, infer metrics, and propose ontologies. Atlan AI Labs reports 690K+ descriptions generated, 87% rated on par or better than human writing, across 50+ enterprise customers. These are the certified definitions an agent needs in canonical form before it can reason accurately.
Context Engineering Studio: Bootstrap, test, and ship context as code, with CI-integrated eval suites that validate context before it reaches production. The same discipline of evaluation Databricks applies to agents in Agent Bricks, Atlan applies to the context those agents consume.
Context Lakehouse: An Iceberg-native, open-format context store that activates via MCP, SQL, and open APIs. Built on open APIs and Iceberg-native formats, context stored in Atlan stays portable, not locked to any vendor’s schema, and reads back into any agent.
| Dimension | Databricks-native context | Atlan Enterprise Data Graph |
|---|---|---|
| Scope | Databricks estate and connected sources | 80+ connectors across warehouses, BI, pipelines, SaaS |
| Grounding | Genie Ontology for Databricks agents | Governed definitions and lineage across the whole estate |
| Generation | Manual and platform-driven | Context Agents: 690K+ descriptions, 87% human quality |
| Validation | Agent Evaluation for agent output | Context Engineering Studio: CI-validated context as code |
| Delivery | MCP and Databricks surfaces | MCP server, SQL interface, open APIs to any agent |
Atlan is a Databricks partner, and the integration runs through the standards both sides already use. AI agents get enterprise context through Atlan’s MCP server, SQL interface, and open APIs, which is the same protocol Agent Bricks uses to reach external tools and data. The connection is additive by design.
Inside Atlan AI Labs and the 5x accuracy factor
See how context engineering drove a 5x accuracy improvement in real customer systems, with a repeatable playbook for getting agents to production.
Download the EbookBetter together: the platform and the context behind it
Permalink to “Better together: the platform and the context behind it”Agent Bricks is a genuine advance in getting agents to production. It handles the 99% of agentic work that sits around the core loop, and the scale numbers (100K+ agents, 1+ quadrillion tokens per year) show enterprises are building on it at volume. Unity Catalog governs it and Genie Ontology grounds it inside Databricks.
The accuracy ceiling for any agent is set by the context it reasons over. When that context is bounded to one platform, or inconsistent across systems, even a well-built agent inherits the gaps. The complete stack pairs the platform with a cross-estate foundation:
- Agent Bricks builds, evaluates, and optimizes agents on governed Databricks data.
- Atlan’s Enterprise Data Graph (80+ connectors) supplies the certified, cross-system context those agents need.
- Context Agents generate the definitions, and Context Engineering Studio tests them before production.
- Context Lakehouse delivers it to any agent via MCP, SQL, and open APIs, portable and open-format.
Agent Bricks plus Atlan means agents with full enterprise context, governed and live, not just lakehouse context. The two are additive. The question for your team is not whether to build on Agent Bricks. It is what foundation of context you are building on.
FAQs about Databricks Agent Bricks
Permalink to “FAQs about Databricks Agent Bricks”-
What is Databricks Agent Bricks?
Agent Bricks is Databricks’ platform for building, evaluating, and optimizing AI agents on governed lakehouse data. Announced as an expanded developer agent platform at the 2026 Data + AI Summit, it brings model choice, agent harness support, MCP-based data access, and Unity Catalog governance into a single surface. (Source: Agent Bricks blog, Databricks, June 2026) -
What was announced about Agent Bricks at Data and AI Summit 2026?
At the June 2026 Data + AI Summit, Databricks expanded Agent Bricks from a quality-focused agent builder into a full developer agent platform. Databricks reported 100K+ agents built and 1+ quadrillion tokens processed per year, and added broader model choice, support for popular agent harnesses, managed agent memory via Lakebase, and MCP-based tool access. (Source: Agent Bricks blog, Databricks, June 2026) -
How does Agent Bricks relate to Genie and Unity Catalog?
Unity Catalog is the governance backbone for agents, tools, and models built in Agent Bricks, while Genie Ontology supplies business context to ground agent reasoning. Genie One and other Genie surfaces are the conversational agents that draw on this foundation. (Source: Databricks Genie One press release, June 2026) -
How many agents have been built on Agent Bricks?
Databricks reported that more than 100,000 agents have been built on Agent Bricks since launch, and that agents on the platform now process more than 1 quadrillion tokens per year. Cited customers include AstraZeneca, 7-Eleven, Fox Corporation, and Block. (Source: Agent Bricks blog, Databricks, June 2026) -
Does Agent Bricks support MCP and third-party agent frameworks?
Yes. Agent Bricks natively supports the Model Context Protocol for secure tool and data access, and supports agent harnesses including LangGraph, CrewAI, Agno, the Claude Code SDK, and OpenAI Agent SDKs, alongside Databricks’ Omnigent meta-harness. (Source: Agent Bricks blog, Databricks, June 2026) -
Why do Agent Bricks agents still need an enterprise context layer?
Agents built in Agent Bricks are only as good as the context they reason over, and most enterprise context lives across Snowflake, dbt, Tableau, Power BI, Salesforce, SAP, and more, beyond Databricks. An enterprise context layer like Atlan unifies governed context across the whole estate and serves it back to those agents via MCP. -
How does Atlan work with Agent Bricks?
Atlan layers on top of Databricks and the rest of the estate, unifying governed business context in the Enterprise Data Graph and delivering it to Agent Bricks agents through its MCP server, SQL interface, and open APIs. Context Agents generate the definitions and Context Engineering Studio tests them before they reach production.
Sources
Permalink to “Sources”- Agent Bricks: Data + AI Summit 2026, Databricks Blog
- Agent Bricks: The governed enterprise agent platform, Databricks Blog
- Databricks Launches Genie One: All-New Agentic Coworker for Every Team, Databricks Newsroom
- What’s new with Unity Catalog at Data + AI Summit 2026, Databricks Blog
- AI governance at Data + AI Summit 2026: what’s new with Unity AI Gateway, Databricks Blog
- Key takeaways from day two of the Databricks Data + AI Summit, SiliconANGLE
- Databricks Bets on Owning the Agentic Data Stack at Data + AI Summit 2026, Moor Insights & Strategy
- Everything Databricks Announced at the DAIS Data + AI Summit 2026, Qubika
- Databricks Summit 2026 Day 2: Agentic AI and Catalog Federation Move From Lab to Enterprise, TechTimes
- Databricks’ new agentic coworker Genie One brings AI automation to every part of the business, SiliconANGLE
