The useful question is what changes for users, developers or buyers, and whether the announcement stays industry context or becomes something people can actually use.

Tool calling allows AI models to recognize when a user's request requires external resources, select the appropriate tool from a predefined set, and generate structured requests—typically in JSON—to invoke that tool. The model does not execute the tool directly but relies on an application layer to handle the call and return results. This mechanism bridges the gap between generative AI, which produces text or code, and agentic AI, which performs multi-step workflows by dynamically selecting and using tools based on context.

The process begins when a user submits a request that falls outside the model's internal knowledge, such as querying sales figures or checking inventory. The model evaluates the intent against available tools, each described with a schema that includes its name, function, and required parameters. For example, a request about revenue would trigger a database query tool, while a weather inquiry would invoke a weather API. The model then generates a structured request for the selected tool, which is executed externally, and the result is returned to the model for synthesis into a natural-language response or confirmation of action.

Tool calling supports complex, multi-step workflows where an agent may need to make sequential calls to different tools. For instance, an agent tasked with preparing a quarterly performance summary might first query a database for metrics, then use a code execution tool to generate a chart, and finally invoke an email API to send the report. This iterative loop enables agents to handle tasks that require planning, data retrieval, computation, and action execution, making them capable of participating in business processes rather than merely providing information.

Tool calling encompasses various types of tools tailored to different use cases. Common categories include data retrieval tools for querying databases or APIs, code execution tools for performing calculations or generating visualizations, action-triggering tools for sending emails or updating records, and integrations with physical systems for monitoring equipment or adjusting settings. These capabilities allow AI agents to operate across digital and operational environments, reducing the need for rigid, rule-based integrations and increasing system flexibility.