Learn AI in Bits podcast

095 - Function Calling: How AI Uses Tools

0:00
5:05
Reculer de 15 secondes
Avancer de 15 secondes


Function calling lets AI models work with external tools, APIs, and business systems. This episode explains the tool-calling loop, how applications define and execute functions, and how OpenAI, Claude, Gemini, and Mistral handle tool requests differently. It also covers schema validation, permissions, and why the application remains responsible for actions.


Voice narration is AI-generated.


REFERENCES (official documentation)


1. OpenAI, Function Calling: https://developers.openai.com/api/docs/guides/function-calling

Tool definitions, function_call and function_call_output, tool_choice, parallel calls, and strict schema mode.


2. Anthropic, How Tool Use Works: https://platform.claude.com/docs/en/agents-and-tools/tool-use/how-tool-use-works

Tool-use contract, client-executed tools, tool_use and tool_result, and the application-managed loop.


3. Anthropic, Handle Tool Calls: https://platform.claude.com/docs/en/agents-and-tools/tool-use/handle-tool-calls

Tool-use identifiers, tool results, and SDK tool runner support.


4. Google, Function Calling with the Gemini API: https://ai.google.dev/gemini-api/docs/function-calling

Function declarations, application-side execution, function results, selection modes, parallel and chained calls.


5. Mistral, Function Calling: https://docs.mistral.ai/inference/conversations/function-calling

The five-step tool-calling sequence and parallel or successive function calls.


Research note: Provider API formats and model capabilities change. This episode describes documented API patterns. It is not a guarantee that every model or SDK supports every option. Check the selected model's current documentation before implementation.


D'autres épisodes de "Learn AI in Bits"