Unified data analytics platform Databricks is taking a significant leap into the enterprise AI space with the launch of Agent Bricks, a new suite of tools designed to simplify and accelerate the development of domain-specific AI agents. Announced at the company’s Data + AI Summit, Agent Bricks is aimed at making the creation and deployment of tailored AI systems more accessible, practical and cost-efficient for businesses of all sizes.
“The fundamental problem we see today is the word agent doesn’t have a well-defined meaning. It’s just some kind of bot that can do something,” Naveen Rao, vice-president, Generative AI, Databricks said. “We’ve tried to make this concept practical for enterprises by identifying common task patterns and building tools around them.”
Agent Bricks is designed to address three key AI agent patterns that Databricks frequently sees in enterprise environments: knowledge extraction, knowledge databases, and domain adaptation. These tools allow users to convert unstructured text into structured formats, build customised chatbots grounded in proprietary data, and fine-tune AI behaviour to specific contexts — all with a user-friendly interface that requires minimal technical expertise.
The flexibility of the platform is enabled by separating the domain (like legal or customer service) from the task (such as querying documents or extracting information). “Our tools customise the entire pipeline based upon the data the user presents to it,” Rao explained. “That’s how we accomplish customisation — by looking at the data itself.”
The new toolset also comes with a built-in feedback mechanism. Rao described it as “very intuitive,” where the agent generates outputs, and users can simply critique them in natural language. “It becomes very similar to how we would train a human,” he said. “You give feedback like, ‘make the summary shorter’ or ‘this is good or bad,’ and the tool adjusts accordingly.”
Agent Bricks also supports ongoing optimisation post-deployment. “AI systems aren’t static. They must improve continually,” Rao said. Databricks offers built-in tools to log inputs and outputs, enabling refinement through both offline and real-time feedback loops.
Early users, including pharmaceutical giant AstraZeneca, have already seen benefits. “They parsed through 400,000 documents to extract structured data without having to write a bespoke solution. In just 60 minutes, they had a working agent,” Rao said, underlining the tool’s rapid time-to-value advantage.
Another feature of Agent Bricks is that it doesn’t commit to a single foundational model. “We’re neutral across model vendors and heavily use open source,” said Rao. “We always aim to use the best tool for the job — whether it’s from OpenAI, Anthropic, or any other vendor.”
Beyond technological prowess, the broader goal is democratisation. “Technology is useless unless more people can access it,” Rao said. “We want to move away from requiring deep technical knowledge to build and use AI systems. Agent Bricks is about unlocking value from your data — quickly and effectively.”
Databricks plans to continue expanding the library of task patterns and orchestration capabilities, offering enterprises a dynamic path into the future of AI.
(The correspondent is in San Francisco at the invitation of Databricks)
