How to Build An AI Agent with Function Calling and GPT-5 | Towards Data Science

Abstraction: Building a web-search AI agent via GPT-5 function calling

Key points:

• Tutorial by Ayoola Olafenwa (2025-10-20) builds a single-tool web-search agent using GPT-5-mini and OpenAI function calling.
• Function calling connects an LLM to external tools (APIs/databases) via a code function plus a JSON Schema describing parameters and when to invoke it.
• Uses Tavily as the web-search tool (1000 free credits/searches on signup); requires OPENAI_API_KEY and TAVILY_API_KEY.
• Uses the OpenAI Responses API (client.responses.create) with previous_response_id to maintain conversation state, reasoning={effort: low}, and text={verbosity: low}.
• Agent decides autonomously: calls web_search for real-time queries (weather, latest iPhone) but answers math/coding directly from pretrained knowledge.
• Tool schema fields covered: type, name, description, strict, parameters, required, additionalProperties=False.

Connections: Openai · GPT-5 · Tavily · AI Agents · Function Calling · Large Language Models

Source: https://towardsdatascience.com/how-to-build-an-ai-agent-with-function-calling-and-gpt-5/