New Databricks tools target successful agentic AI development
With enterprises struggling to build agents, new Agent Bricks features address accuracy and governance to help move development projects from pilot to production.
Databricks' suite for building AI agents is getting an update.
Beginning its Week of AI Agents, the vendor on Monday made agent quality and observability capabilities in Agent Bricks, aimed at improving agents' accuracy, generally available. In addition, Databricks unveiled other features, such as a Model Context Protocol (MCP) catalog to provide governance of agents and capabilities that enable access to unstructured data that are in public preview.
Agent Bricks, unveiled in beta testing in June, was designed to make it easier for customers to move development projects from pilot to production.
Although many data management and AI vendors have introduced tools to simplify building AI models and applications, including agents, an overwhelming majority -- estimated as as high as 80% -- of AI projects never make it to production. Agent Bricks features automate many of the steps involved in developing and deploying agents to try to make building agents more successful.
In September, as part of a $100 million partnership with OpenAI, Databricks made OpenAI's models natively available in Agent Bricks as well as the broader Databricks Data Intelligence Platform.
Given that the new Agent Bricks features address concerns, such as accuracy and governance, that prevent many AI development projects from making it into production, they are valuable additions, according to William McKnight, president of McKnight Consulting.
"The new capabilities are a significant update designed to instill confidence in moving AI agent projects from pilots to secure production by focusing on ensuring the AI is governed, open and accurate," he said. "A full agent lifecycle is covered."
Based in San Francisco, Databricks was one of the pioneers of the data lakehouse format for storing data.
Striving for success
Like many of its data management peers, Databricks ventured into AI development after OpenAI's November 2022 launch of ChatGPT marked a significant improvement in generative AI (GenAI) technology and sparked a surge of interest in AI development that continues to increase.
The new capabilities are a significant update designed to instill confidence in moving AI agent projects from pilots to secure production by focusing on ensuring the AI is governed, open and accurate. A full agent lifecycle is covered.
William McKnightPresident, McKnight Consulting
Enterprises mainly built GenAI chatbots following ChatGPT's release. By mid-2024, enterprises had shifted their development focus to agentic AI. Agents, unlike chatbots that require human prompts before acting, can be trained with contextual awareness and reasoning capabilities that enable them to autonomously take on certain tasks.
Agents, however, are hard to build. And because they are designed to act autonomously, they must work properly so they don't do things that could harm an organization.
Agent Bricks is Databricks' suite for simplifying agentic AI development.
The update, which includes its first generally available feature and focuses on ensuring accuracy, enforcing governance and enabling access to data sources, is significant, according to Devin Pratt, an analyst at IDC.
"Collectively, these updates help organizations move agents from pilot to production with greater control and trust," he said. "This is about making enterprise agents trustworthy, accurate, governed and flexible on the data organizations already control.”
Specific features of the Agent Bricks update include the following:
MLflow for Agent Quality and Observability to continuously monitor and evaluate agents to improve their accuracy.
AI Gateway, a governed interface for managing agent connections to models such as OpenAI's GPT-5, Google's Gemini, Anthropic's Claude Sonnet and open source models.
MCP Catalog in Marketplace to provide similar governance and management of agent connections to other external tools and data sources.
MCP support in Agent Bricks Multi-Agent Supervisor enabling users to coordinate multi-step workflows involving specialized agents.
ai_parse_document, a SQL function that enables extract content from unstructured data such as documents and tables to give agents greater context than structured data alone provides.
MLflow for Agent Quality and Observability is generally available while the other features are in either public preview or beta testing.
The problems many organizations have when moving agentic AI projects into production were the catalyst for Databricks adding the new Agent Bricks features, according to Craig Wiley, senior director of product for AI and machine learning at Databricks.
Wiley noted several issues that hinder agentic AI development, such as lack of confidence in agent quality and accuracy, agents locked into a single model provider, and security and governance concerns.
"The impetus for releasing these new Agent Bricks capabilities was to address these challenges by … helping companies choose the best model for their task and extending governance and security controls enterprises already trust," Wiley said.
While the Agent Bricks suite is designed to foster successful agentic AI development, the new features that could benefit Databricks users the most are the MCP Catalog in Marketplace and ai_parse_document, according to McKnight.
"They directly solve core challenges related to governance, security and data grounding that typically cause AI agent projects to stall in the pilot phase," he said.
Pratt likewise noted the significance of MCP Catalog in Marketplace but also highlighted MLflow for Agent Quality and Observability.
"MLflow for Agent Quality and Observability stands out for making evaluation and continuous improvement part of the standard workflow, which is essential for regulated or customer-facing use cases," he said.
While Agent Bricks is designed to help customers more successfully develop agents, Databricks is not the only data management vendor providing an environment aimed at making it easier for users to build agentic AI tools. For example, Databricks rival Snowflake provides Cortex Agents. In addition, Teradata, Informatica and tech giants AWS, Google Cloud and Microsoft are among others also providing agentic AI development capabilities.
Databricks, however, differentiates itself by unifying data governance, model control and agent evaluation and integrating those capabilities with a data lakehouse, according to Pratt.
"This gives Databricks an edge in delivering governed, data-centric AI operations while remaining consistent with peers in supporting flexible and well-orchestrated AI development," he said.
McKnight, meanwhile, noted that Databricks' focus on directly addressing problems related to advancing AI development projects beyond the pilot stage is unique.
"Databricks is actively differentiating itself … by focusing its new capabilities on the important core enterprise hurdles of governance, accuracy and production scale," he said.
Other vendors have advantages when it comes to analytics and integration capabilities, McKnight continued.
"Thus, Databricks is positioning itself as the dedicated, governed platform for complex AI execution, leveraging its unique strength in combining data, machine learning and unified governance," he said.
Looking ahead
Databricks plans to make it easier and safer for enterprises to develop agentic AI tools trained on their proprietary data, according to Wiley.
"This will include improving AI agent quality, accuracy and governance so they can trust AI," he said.
Databricks is wise to focus on governance, according to Pratt. In addition, he noted that its emphasis on transparency and open model ecosystems is providing customers with desired capabilities.
"Databricks is leaning into what enterprises need most right now -- confidence, control and continuity," Pratt said.
However, despite its focus on trying to make AI development more successful, Databricks still has issues related to the ease-of-use of its platform and the clarity of its pay-as-you-go pricing, according to McKnight. To appease customers and potentially attract new ones, he suggested that the vendor address those areas.
"Databricks must address its user interface to make onboarding easier and attract customers seeking both AI and traditional business intelligence," McKnight said. "Addressing cost opacity and total cost of ownership is also critical, as Databricks' pricing is confusing. Competitors are offering [cost] savings, driven partly by Databricks' performance in integration workloads that causes costs to rise."
Eric Avidon is a senior news writer for Informa TechTarget and a journalist with more than 25 years of experience. He covers analytics and data management.
