AI agents can feel magical from the outside, but under the hood, they are built from a few practical building blocks. In this talk, we will build a minimal AI Agent SDK from scratch in Python and use it to understand how those pieces fit together.
We will start by creating a small Python package and designing it to work across multiple LLM providers, including OpenAI, Anthropic, and others. From there, we will explore basic agent concepts such as system prompts, context, and messages, and gradually add support for tool calls, MCP, and skill files. By the end of the talk, attendees will understand the core structure of an agent SDK and feel more confident building or modifying one for their own needs.