Why do we import 100MB of frameworks to run a 50-line LLM reasoning loop?

Short summary

This post argues that building AI agents doesn't require heavy orchestration frameworks like LangChain or AutoGen. It breaks down agentic design into three core pillars: State (message memory), Schema (tool definitions), and Loop (reasoning cycle). The author provides a minimal Python implementation using the OpenAI SDK that demonstrates the complete pattern while maintaining full visibility into debugging and control.

• •Heavy frameworks like LangChain/AutoGen add complexity without necessity
• •All agents need three components: state (memory), schema (tools), reasoning loop
• •Minimal vanilla implementation provides better debugging and full control

Generated with AI, which can make mistakes.