Databricks Enhances AI Agents with Tool Calling Capabilities

Tool calling enables AI models to interact with external tools, APIs, and systems, transforming basic chatbots into capable AI agents that can take real-world action. This process involves the model recognizing when a tool is needed, selecting the appropriate one, building a request, processing the response, and delivering a result or executing an action.

The system supports various tool calling patterns, including information retrieval (e.g., database queries, document search, RAG) and code execution for calculations, data transformation, or running analytical scripts. This allows AI agents to perform a wide range of tasks beyond their inherent knowledge.

Databricks Agent Bricks offers a governed platform for building tool-calling agents grounded in enterprise data. It provides native support for the Model Context Protocol (MCP) and Unity Catalog governance, ensuring secure and reliable agent development.

The tool calling process involves a structured loop where the AI model, not executing tools directly, generates structured requests (often JSON) that an application layer fulfills. The external tool's response is then fed back to the model to synthesize a natural-language answer or confirm an executed action, enabling multi-step workflows.

Tool calling is the mechanism that separates static chatbots from functional AI agents by allowing models to perform actions beyond text generation. Gartner predicts a significant increase in enterprise applications embedding task-specific AI agents, highlighting the growing importance of reliable tool calling for production-grade value and reduced engineering overhead.