Databricks Pilots 'Agent Bricks' Assistant
Databricks is piloting a new "Knowledge Assistant" internally, nicknamed "Agent Bricks." The context-aware copilot is designed to help users with documentation, troubleshooting, and code generation specific to the Databricks platform. The goal is to reduce onboarding time and improve platform reliability in complex environments.
The "Agent Bricks" initiative is powered by Databricks' Mosaic AI and is part of a broader strategy to embed generative AI across its Data Intelligence Platform. This follows the company's high-profile, $1.3 billion acquisition of generative AI platform MosaicML, known for its MPT large language models. The entire MosaicML team, including its research division, was integrated into Databricks to accelerate the development of AI models that organizations can build, own, and secure with their own data. Agent Bricks is designed to automate the heavy lifting in creating domain-specific AI agents. Users can define tasks in natural language, and the system automatically tests various models and configurations to optimize for both performance and cost. It leverages techniques like synthetic data generation to improve agent quality without manual data labeling. The system is tightly integrated with the Databricks ecosystem, including MLflow for evaluation and Unity Catalog for governance. This integration is crucial for regulated industries like healthcare, providing a unified platform for managing data, analytics, and AI with built-in security. The platform aims to reduce the cost of training and using large language models from millions to thousands of dollars. This push into AI agents is supported by a series of strategic acquisitions beyond MosaicML. Databricks has also acquired Okera for data governance, Arcion for data replication, and Neon, a serverless Postgres company, to enhance the capabilities of its Lakehouse platform for AI-driven applications. For developers, Databricks has been rolling out AI Functions that allow analysts to apply generative AI models directly within their SQL queries, simplifying tasks like sentiment analysis and text summarization. This functionality is powered by models like Mixtral-8x7B and is designed to make AI more accessible to users who are more comfortable with SQL than with Python. The broader vision is to create a "Lakebase" architecture, a new category of database designed for developing data-intensive applications and AI agents. This strategy positions Databricks to compete in the evolving landscape of AI-native applications, where AI agents can autonomously perform complex tasks by interacting with enterprise data systems.
