What Early Access to Databricks’ Agent Bricks Taught Us — and Why It Matters for Your Business

A conversation with our CTO about what Analytics8 learned during early access to Agent Bricks—and how those lessons apply as Databricks expands its platform for building and governing enterprise agents.

Databricks introduced Agent Bricks at the 2025 Data + AI Summit and has since expanded it into a comprehensive platform for building, deploying, and governing enterprise agents. Before Agent Bricks had a public launch, user interface, or even a name, Analytics8 was one of a small number of organizations selected to participate in the private preview. That gave us months to work directly with Databricks product and R&D teams—and to test, challenge, and learn from the technology before it reached the broader market.

1. What is Agent Bricks, and what value does it bring to businesses?

Agent Bricks is Databricks’ platform for building, deploying, optimizing, and governing enterprise agents. It gives developers the flexibility to use different models, frameworks, tools, and data sources while Databricks provides the underlying infrastructure, governance, evaluation, and control. Organizations can build agents that work across structured and unstructured enterprise data without assembling every component from scratch.

The easiest way to think about it? It’s like ChatGPT — but with secure access to your internal content. That includes policies, presentations, documentation, white papers, and more. By grounding agents in governed enterprise data and context, organizations can generate responses based on what their business knows—not just what a general-purpose model knows.

Historically, setting something like this up required a lot of behind-the-scenes plumbing: vector databases, embedding models, chunking strategies, front-end tools, role-based access layers, feedback loops, countless optimizations, and more. Agent Bricks abstracts all of that.

Databricks manages much of the underlying infrastructure, allowing teams to spend less time assembling technical components and more time preparing data, evaluating quality, establishing governance, and integrating agents into business workflows.

That’s the biggest shift: it moves AI implementation from something that takes months to something you can stand up in hours.

And the speed doesn’t come at the expense of governance. Because Agent Bricks sits within the Databricks lakehouse environment and is governed by Unity Catalog, you get the same enterprise-grade access controls, auditability, and security you’d expect from the rest of your data estate.

From our perspective, the biggest value is this: we no longer have to spend time building infrastructure — we can focus on helping clients with data quality, building data pipelines, and ultimately, connecting data to outcomes. That means aligning AI use cases to business goals, shaping a governance strategy, and preparing teams for adoption, instead of getting stuck in the technical setup.

2. Why was Analytics8 selected for early access?

Databricks product management knew we could bring a broad range of real-world use cases — and meaningful feedback — to the table.

Through conversations with solution architects, engineers, and product leaders, it was clear they wanted early input from partners who weren’t just experimenting but who had active, complex needs for internal and client-facing AI use cases. We had both. We brought everything from HR assistants to complex research bots built on top of multimodal documentation.

That variety made our feedback valuable, and the partnership mutually beneficial. They got insight into how the tools were performing across use cases, and we got direct visibility into the roadmap and the chance to influence what would launch (and what wouldn’t) in the early product direction.

3. How did we use the opportunity, and what did we learn?

Being part of the private preview wasn’t just about testing — it gave us space to think critically and challenge long-held assumptions about how unstructured data fits into a modern AI strategy.

Most companies treat unstructured data as something that lives outside the analytics stack — in SharePoint folders, Google Drives, or scattered file shares. And while data lakehouses have always technically supported unstructured data, that support has rarely translated into actual lakehouse use cases.

Working with Agent Bricks forced us — in the best possible way — to confront that gap. If unstructured data is now powering AI assistants, it can’t live in the shadows anymore. t must be governed, discoverable, and accessible to the agents that need it. Whether organizations ingest that content into the lakehouse or connect to it through external knowledge sources, they need clear access controls, metadata, quality standards, and lifecycle management.

That shift completely changed how we approach data readiness and governance. We rethought how data pipelines should work. We built accelerators to help our clients ingest and catalog unstructured content quickly and securely. And we developed strong points of view on where unstructured data belongs in the broader architecture.

The other big unlock was time. Because we weren’t racing to catch up after a public launch, we got to move deliberately. We worked directly with Databricks’ R&D and product teams, helped influence the roadmap, and had the opportunity to think, not just react. That gave us clarity on how to approach Agent Bricks strategically, not just tactically.

The result? We’re not just familiar with the tool — we’re already helping clients make architectural decisions, define use cases, and avoid common mistakes. We didn’t just use the opportunity. We built on it.

4. How does that experience help us guide clients through their AI strategy — even if they’re just getting started?

We’ve already done the thinking, the testing, and the building, so clients don’t have to start from zero.

Because we were part of the preview, we’ve had time to refine our point of view on where these tools fit in an overall AI strategy. And we’ve started building assets and accelerators to make implementation faster and more repeatable. We’re not guessing at what works — we’re already applying what we’ve learned.

More importantly, we’ve seen where the real work still lies. Much of the underlying platform infrastructure is now handled by Databricks. What clients need help with is identifying the right use cases, getting their data in shape, and rethinking how AI fits into business workflows. That’s where we’re focused.

5. Most folks are just now learning about Agent Bricks. What’s something important they’ll miss if they don’t partner with someone who’s already been in the weeds?

The complexity under the hood — and what it takes to operationalize this.

When we first got access, there was no user interface. We were deploying from the command line, running scripts, and debugging raw outputs. That gave us a deep understanding of how the system works — and where the edge cases live.

That kind of exposure helps us move faster and troubleshoot better. It also helps us advise clients with confidence. Tools like this are powerful, but they’re not magic. You still need to understand what’s happening under the surface — or work with a partner who does.

6. How has the Agent Bricks vision evolved since the private preview?

One of the developments we were most excited about during the preview was the ability to unify access to structured and unstructured knowledge through supervisor agents. That vision is now becoming real. Agent Bricks Supervisor Agent can coordinate agents and tools—including Genie Spaces, Knowledge Assistant agents, and MCP servers—to answer more complex questions through a single experience.

7. How has this experience shaped Accelr8, our AI-enabled delivery framework?

The approach we brought to Agent Bricks is reflected in Accelr8, our AI-enabled delivery framework. Accelr8 combines AI agents, automation, reusable assets, and senior practitioner oversight to help data and AI projects move faster without sacrificing quality, governance, or business alignment.

The unstructured-data pipeline we developed during the Agent Bricks preview is one example. Instead of rebuilding foundational components for every client, Accelr8 lets us apply proven patterns and adapt them to each organization’s environment and goals.

Accelr8 extends beyond Agent Bricks and Databricks, supporting multiple platforms and the full delivery lifecycle—from discovery and design through build, testing, documentation, and rollout.

Learn more about Accelr8.

Key Takeaways

• Agent Bricks is Databricks’ platform for building, deploying, optimizing, and governing enterprise agents across structured and unstructured data.
• Analytics8 was part of a small group selected for early access, which allowed the team to work directly with Databricks’ product and R&D teams before the public launch.
• This experience led Analytics8 to rethink how unstructured data fits into modern AI strategies by treating it as a governed and queryable asset within the lakehouse.
• During the preview period, Analytics8 built real-world use cases, developed accelerators, and contributed feedback that shaped the product’s early product direction.
• Today, Analytics8 helps clients focus on aligning AI tools with business goals, preparing data, and embedding AI capabilities into day-to-day workflows.
• With deep experience in the underlying systems, Analytics8 can deploy solutions more quickly, troubleshoot more effectively, and guide strategic implementation decisions.
• The team created an accelerator that moves unstructured content from platforms like SharePoint into the Databricks lakehouse, applying governance through Unity Catalog to make the data usable for AI applications.
• Supervisor Agent can coordinate agents, Genie Spaces, and other tools to answer questions that span multiple data sources and systems through a unified experience.