Session brief
Function Calling is All You Need — Full Workshop, with Ilan Bigio of OpenAI
Overview
This workshop explores the power and versatility of function calling in AI models, arguing that it is a fundamental capability for building advanced AI applications. The talk traces the evolution of language models from simple text completion to instruction following and finally to sophisticated tool use. It demonstrates how function calling enables AI to interact with external systems, fetch data, take actions, and manage complex workflows, forming the backbone of agentic behavior.
Who should watch
- AI engineers and developers looking to integrate AI models with external tools and APIs.
- Product Managers and builders interested in understanding the core mechanisms behind agentic AI.
- Anyone curious about how AI models can perform actions beyond generating text, such as data retrieval or state management.
- Developers exploring advanced AI patterns like memory, delegation, and asynchronous operations.
Key takeaways
- Function calling has evolved from early text completion models to instruction-following models and now to general-purpose tool use, significantly expanding AI capabilities.
- The core purposes of function calling are fetching data (e.g., from APIs, databases) and taking actions (e.g., managing state, executing workflows).
- When implementing function calling, it's crucial to apply software engineering best practices, including clear function descriptions, parameter explanations, and using enums/object structures to prevent invalid calls.
- Agents can be conceptualized as loops that specify tools, call a model, handle tool calls, append results, and repeat until a task is complete.
- Memory can be implemented simply as a list or more complexly with retrieval and semantic similarity, allowing agents to retain and utilize past information.
- Delegation allows an agent to hand off tasks to other agents or call more powerful models, enabling complex problem-solving.
- Asynchronous operations are key for non-blocking tasks, allowing AI to perform multiple operations concurrently, such as parallel API calls or background processing.
- Advanced patterns like task creation and checking enable agents to manage background processes, providing a more responsive user experience.
Notable quotes
*Function calling is all you need. It's a talk workshop. There's going to be um a lot of coding.*
*Agents are just loops. How rag workflows and more are just function calls delegation and asynchrony.*
*The model tells you what it wants to do with that function but it doesn't actually do it. This is one of the like a big sticking point function calling. It doesn't actually use the function itself.*
Unofficial community note. Prefer the recording for nuance.