Agent Bricks: Databricks’ platform that wants to solve the biggest bottleneck in enterprise AI
Artificial Intelligence for businesses is no longer a distant promise — it has become a very real, day-to-day pressure.
But there is a detail that many people still underestimate: building an AI agent is the easy part. The basic pattern is already well known — a model connected to tools, reasoning and executing actions. Setting up that loop is no longer the problem.
The real challenge begins when that agent needs to work with real enterprise data, respecting permissions, understanding business context, and operating with real-world consequences — not in a test environment.
An agent that analyzes credit applications and applies a financial institution’s internal underwriting policies is only valuable because it understands that business’s context — not because it uses a more sophisticated model or a specific framework. It is the depth of the connection with customer records, operational systems, internal policies, and institutional knowledge that makes these agents truly useful.
This is exactly where most solutions hit a wall: they deliver pieces, but not a complete platform.
That is precisely why Databricks developed Agent Bricks, an enterprise platform built from the ground up to create, deploy, and govern agents that operate on your organization’s data, end to end. The platform unifies access to models, execution, governance, and context so that teams can run agents reliably in production.
In this article, you will learn what makes this platform different, what is new, and why governance is no longer a technical detail but the heart of any serious AI-in-production strategy. 🚀
What is Agent Bricks and why it matters now
Agent Bricks is Databricks’ answer to a question that corporate technology teams have been trying to answer for a while: how do you bring artificial intelligence agents into real production without sacrificing control, security, and traceability? The platform was not created to compete with rapid prototyping tools or experimentation environments. It was designed for the real world of large organizations, where the wrong data accessed at the wrong time can mean a compliance violation, a flawed decision, or even a serious regulatory issue. That completely changes the scale of the problem that needs to be solved.
The core value proposition of Agent Bricks is to integrate, in a single enterprise platform, everything an AI agent needs to exist with quality in a corporate environment: access to enterprise data with granular permission controls, tools for building and orchestrating the agent’s reasoning flows, continuous evaluation infrastructure, and at the center of it all, a robust governance layer. It is not a collection of standalone APIs. It is an architecture designed so that the pieces work together, reducing friction between data, engineering, and business teams.
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
Thousands of organizations across sectors like financial services, retail, healthcare, and technology have already deployed agents in production and at scale on Agent Bricks. Companies such as Workday, Virgin Atlantic, Zapier, EchoStar, and AstraZeneca are among those using the platform. The use cases are varied: teams building agents that deliver continuous market analysis for hundreds of analysts, orchestrate workflows across supply chain, procurement, and R&D, automatically resolve employee requests involving complex tasks, and detect anomalies in marketing campaigns before ad spend is wasted.
It is also worth understanding the context in which this platform arrives. The enterprise AI agent market has grown rapidly over the past two years, but most available solutions still require companies to build their own control infrastructure on the side. The result is improvised security layers, audit difficulties, and agents that work great in development but stall when it is time to scale. Agent Bricks was built specifically to close that gap, bringing inside the platform what previously had to be hand-built by each team.
The three pillars that define the platform
Databricks structured Agent Bricks around three characteristics that, according to the company, are what separates a real platform from a disconnected set of tools.
Open and multi-AI
To build truly useful agents, teams need to work with multiple model providers and frameworks. Agent Bricks natively supports frontier models and popular code agents like Cursor, Codex, and Claude Code through a unified API, with built-in routing, automatic fallback, and cost optimization. It is also possible to build and deploy agents with frameworks like LangGraph and OpenAI Agents SDK.
This allows teams to swap models or integrate external agents without having to rebuild their systems. An interesting data point: 63% of the platform’s customers already route tasks across two or more model families, ensuring that agents stay flexible and resilient as models evolve.
Unified governance
Most platforms only govern the agent itself — which tools it can call and what permissions it has. Agent Bricks governs the agent and everything it interacts with in a single system. With Unity Catalog and the AI Gateway, access to data, models, and external MCPs is managed and observed in one place, with identity enforced end to end.
Agents inherit the user’s identity through on-behalf-of token passing. In practice, this means the agent can only access what the user is authorized to use. Whether querying the lakehouse or calling external APIs, the same permissions, audit trails, and routing rules apply to every interaction.
Accurate because it understands business context
An agent’s accuracy depends on much more than model quality. Agent Bricks uses Unity Catalog metadata — including schema, business definitions, lineage, permissions, and data quality signals — to improve how agents reason and act. This context is incorporated directly into retrieval and planning, delivering, according to Databricks, 70% more accuracy than standard RAG and a 30% improvement in multi-step workflows.
For structured data, Genie Spaces leverage the semantic layer so that agents reason about business definitions, not raw column names. This ensures that agents return answers aligned with how the business actually operates. 🎯
Governance as a core structure, not an afterthought
One of the most significant aspects of Agent Bricks is how governance is positioned within the platform’s architecture. In many enterprise artificial intelligence solutions, governance shows up as a layer added after the fact, almost like a compliance plug-in. In Agent Bricks, it sits at the core of the design. This means that every agent created on the platform is born with built-in traceability, native access controls for enterprise data, and mechanisms that allow auditing what the agent did, why it did it, and based on what information it reached that conclusion. For sectors like finance, healthcare, and legal, this is not a differentiator — it is a baseline requirement to put anything into production.
Governance within Agent Bricks leans heavily on the Unity Catalog infrastructure, which is already Databricks’ data cataloging and control system. This means that the permissions an organization has already defined for its data — who can access what, in what context, and at what level of detail — are automatically respected by agents operating on the platform. An HR agent cannot access financial information because the catalog already knows that data is out of scope for that agent. This integration eliminates an entire layer of manual configuration that, when done externally, is typically the biggest source of security failures in enterprise AI projects.
As William Acosta, Head of Agentic AI Engineering at EchoStar, put it: with Agent Bricks, the company is not building isolated AI projects but rather an enterprise AI mesh, where interoperability, identity-based security, and governance were designed from day one, making agents behave like any other mission-critical system.
Beyond that, the platform offers data lineage traceability applied to agents. This means it is possible to know exactly where each piece of information the agent used to make a decision came from, which version of the data was queried, and whether any transformations occurred along the way. For compliance and audit teams, this represents a significant shift in how they work with AI in production. Instead of trying to reconstruct the agent’s logic after something goes wrong, the organization already has the complete record available from the very first moment. This is governance functioning as infrastructure, not as bureaucracy. 🔍
What is new on the platform
Databricks announced general availability (GA) for Document Intelligence and Custom Agents, along with new capabilities across the platform, including the AI Gateway. The updates span three main areas:
Multi-AI and agent orchestration
- Custom Agents on Apps (GA) — Allows building and deploying agent applications with any model or framework, with full lifecycle support and serverless compute. Native integration with Lakebase provides memory, conversation history, and state for long-running workflows.
- Supervisor Agent (GA) — Orchestrates multiple agents and tools in a single workflow. You define the task and connect your systems, and the supervisor coordinates execution across models and tools.
- Web Search in Foundation Model API — Grounds agent responses with real-time information from the web, using native search capabilities from providers.
Governed access to tools, models, and data
- AI Gateway — A unified layer for managing and governing access to models, code agents, and now tools connected via MCP. It enforces identity, permissions, and observability on every interaction. It now also includes guardrails for detecting and mitigating risks such as PII exposure, unsafe content, prompt injection, data exfiltration, and hallucinations, with customizable options for different security needs.
- Managed OAuth MCP Connectors — Connects external services like GitHub, Atlassian, and Glean as governed tools. Credentials are managed centrally so agents can access systems without exposing secrets.
Enterprise context that makes agents accurate
- Document Intelligence (GA) — Extracts and structures data from unstructured documents like contracts, invoices, and reports, turning PDFs into searchable knowledge without custom pipelines.
- Knowledge Assistant (GA) — Automatically ingests enterprise documents and makes them available to any agent, with retrieval that incorporates system context, metadata, and user restrictions.
- Agent Mode in Genie Spaces — Evolves from single-turn Q&A to multi-step reasoning and analysis over your data, allowing agents to plan, explore, and answer complex business questions.
- CLEARS Framework for agent quality with MLflow — Evaluates agents across six dimensions: correctness, latency, execution, adherence, relevance, and safety, with a standardized framework in MLflow for production quality.
The experience of Zapier illustrates the practical impact of these capabilities well. According to Alvaro Martin, senior data engineer at the company, Agent Bricks provides a structured way to coordinate multiple data intelligence endpoints into a single system. Instead of manually coding routing logic, it is possible to guide how the agent prioritizes Genie and governed data in Unity Catalog through clear instructions, which makes it much easier to build an internal data query experience that is flexible and reliable as it evolves. 🤝
Why enterprise data changes everything
There is a fundamental difference between an AI agent trained on public data and one that operates on an organization’s enterprise data. The latter has access to the real context of the business: transaction history, internal policies, customer base, active contracts, product structure, and all the intelligence accumulated over years of operations. That context is what transforms a generic agent into a genuinely useful tool. But it is also what creates the biggest risk, because that data is sensitive, frequently regulated, and in many cases, the organization’s primary competitive asset.
Agent Bricks was built starting from that premise. The platform does not treat enterprise data as just another input that flows into the agent and disappears. It treats it as a managed resource, with a lifecycle, permissions, and traceability. This means that when an agent queries the customer database to personalize a response, there is a record of that access, tied to the conversation context, the agent that made the query, and the result that was generated. That level of control is what allows an organization to maintain its security posture even after opening its data for agents to work with.
On top of that, the platform offers native support for what Databricks calls proprietary data as a competitive advantage. The idea is that agents are not just better because they use a more powerful language model, but because they have access to information that no public model has. That combination — an artificial intelligence model with the organization’s exclusive data, operating within a robust governance layer — is what Databricks bets will define which companies actually manage to extract strategic value from AI in the years ahead. It is a vision that makes sense, especially for organizations that have already invested years building their data assets and now want to put that investment to real work. 💡
The ecosystem behind the platform
One thing that sets Agent Bricks apart from other initiatives in the market is the partner ecosystem forming around the platform. Databricks has been expanding this network with integrators and technology companies that extend agent capabilities beyond the native environment. Partners already announced include names like Accenture, Capgemini, Deloitte, EY, Infosys, Wipro, LTIMindtree, Slalom, and Tiger Analytics on the consulting and implementation side, plus technology companies such as Arize, Collibra, Dataiku, Glean, LlamaIndex, Monte Carlo, Qlik, Retool, and Sigma Computing.
This ecosystem matters because, in practice, no enterprise AI platform works in isolation. Companies use dozens of tools and services in their daily operations, and agents need to connect to those systems to be truly useful. Having partners that are already building native integrations with Agent Bricks significantly reduces the time between deciding to adopt the platform and the moment agents are actually in production, generating value.
The challenge is no longer the agent loop
Databricks makes it clear that today’s challenge is no longer building the agent loop. What really needs to be solved is everything around it: identity that cannot be bypassed, credentials that do not leak, model routing that avoids lock-in with a single vendor, business context that makes results accurate, and observability that shows what each agent did and why.
Most platforms focus on pieces of that system. Agent Bricks aims to bring it all together in one place.
Agent Bricks represents an important mindset shift in the enterprise artificial intelligence market: the idea that governance, data, and agents are not separate problems that need to be solved by different teams, but parts of a single solution that must work in an integrated way to deliver real value.
For teams evaluating how to evolve their AI-in-production strategy, understanding how platforms like Agent Bricks solve the problem end to end is the most honest starting point for making good decisions. It is not about which technology is the most cutting-edge, but about which approach actually works when the data is real, the risks are concrete, and the results need to be explainable. 🎯
