Member-only story
Build Intelligent Tool-Using Agents Without Frameworks Using Python, Pydantic, and the ReAct Pattern
13 min readJust now
–
Press enter or click to view image in full size
Non-members link
If you ask me what’s the best way to learn AI Agents in 2026, I will tell you it is definitely by building them yourself from scratch. That’s not only important for learning purposes, but if you’re going to build a production-grade AI agent that needs to be highly efficient, personalized and robust, then building one from scratch is your best option. For instance, all of the coding agents you can find (e.g., Claude Code, Codex, Cursor, etc.) are built with custom architectures specific to their products.
Now don’t get me wrong, frameworks like LangChain, LangGraph,…
Member-only story
Build Intelligent Tool-Using Agents Without Frameworks Using Python, Pydantic, and the ReAct Pattern
13 min readJust now
–
Press enter or click to view image in full size
Non-members link
If you ask me what’s the best way to learn AI Agents in 2026, I will tell you it is definitely by building them yourself from scratch. That’s not only important for learning purposes, but if you’re going to build a production-grade AI agent that needs to be highly efficient, personalized and robust, then building one from scratch is your best option. For instance, all of the coding agents you can find (e.g., Claude Code, Codex, Cursor, etc.) are built with custom architectures specific to their products.
Now don’t get me wrong, frameworks like LangChain, LangGraph, and LlamaIndex can also be useful for standard tasks, like RAG or automated workflows. The point is that you need to be aware of the capabilities and limitations of any library before you decide to use it on a real complex task.
I still use LangGraph a lot, but mostly in prototyping. It is really great for creating a demo, or for teaching design patterns and agentic architectures.
In this and the next post, I’ll show you how to build step-by-step an AI agent with essential…