You've seen the theory: workflows vs. agents, the types of agents, the patterns for coordinating them. This module is where that theory turns into code. Building agents is the hands-on skill of wiring an LLM up to a tool-calling loop — describing functions, letting the model choose one, executing it, feeding the result back, and repeating until the task is done.
Both lessons here are full, project-based courses rather than short explainers. The first takes you from LLM fundamentals through the ReAct pattern to a working multi-agent system, building four distinct agent personalities from scratch and closing with a teardown of a real production agent framework. The second strips everything down to one focused build: a CLI coding agent, written in plain Python against the Gemini API, that can read a codebase, find a bug, and fix it — no framework, just the agentic loop itself.
Watch the lessons in order, then continue on to Security & Ethics.
Mumshad Mannambeth's full-length course goes from LLM core concepts and tokens, through workflows vs. agents and the ReAct pattern, to building four agent personalities — an orchestrator, a research specialist, a memory manager, and a coding agent — plus guardrails, structured outputs, and a teardown of the OpenClaw agent framework.
Boot.dev's Lane Wagner builds a toy version of Claude Code from scratch using the free Gemini API — function declarations, a sandboxed working directory, and the agent loop that lets the model read, write, and run code until a real bug is fixed.