Tool Call
What is a tool call?
A tool call is a specific action where an AI agent requests to execute a tool—such as calling an API, accessing a database, or running code—during an interaction with a user. Tool calls represent the intermediate steps between user input and final LLM responses, capturing how the agent gathers information or takes action to fulfill a request.
Tool calls are always bidirectional: the agent makes the call and receives a response that informs its next actions.
How are tool calls captured?
Tool calls are recorded as part of a trace—the detailed log of an AI interaction. A complete trace includes:
- User input
- System prompts
- Tool calls made by the agent
- Results returned from those tool calls
- Intermediate processing steps
- Final LLM responses
This observability makes tool calls debuggable and helps teams understand how their AI agents are behaving. When an agent doesn't perform as expected, teams can examine the trace to see which tools were called, what data was returned, and how that influenced the final response.
What distinguishes modern agents from early chat interfaces?
Tool calling capability represents a significant evolution in AI systems. Early chat interfaces (circa 2023) could only generate text responses. They couldn't look up information in databases, make API calls, or execute code.
Modern agents (2025 and beyond) can autonomously decide to invoke tools based on user needs. For example, a customer interview coaching agent might call a "Student Transcript Lookup" tool to retrieve a student's interview history before providing feedback, then call an "Update Student Transcript History" tool to save its coaching response.
This ability to interact with external systems and data—rather than just generating text—makes agents substantially more capable and useful.
Learn more:
- Building My First AI Product: 6 Lessons from My 90-Day Deep Dive
- Turning Disruption into Opportunity: The Stack Overflow AI Story with Ellen Brandenberger
Related terms: