Building AI Agents
Give LLMs tools, memory and the ability to act autonomously
Build real agents: tool calling, the ReAct loop, agent memory, planning, error recovery and production agent patterns. Move from single-shot completions to autonomous systems that act.
Who this series is for
This series is for engineers who are comfortable with LLM APIs, structured outputs and basic function calling, and now want to build systems where the model decides what to do next. Agents are easy to demo and hard to make dependable. The emphasis throughout is on control: bounding what an agent can do, detecting when it is stuck, and recovering from tool failures.
What each chapter covers
- Tool Calling covers designing tool schemas the model can use correctly, validating arguments before execution, and returning errors in a form the model can recover from.
- Agent Loops covers the reason-act-observe cycle, stopping conditions, step and cost budgets, and the specific ways loops fail in practice, such as repeated calls and drifting goals.
- Agent Memory & Planning covers short-term and long-term memory, what belongs in context versus in a store, and when explicit planning helps versus when it adds latency for nothing.
How to approach it
Build one small agent with two or three tools and keep extending it through the series, rather than starting fresh each chapter. Log every step of every run. Reading traces of your own agent's failures is the fastest way to understand why the guardrails in each chapter exist. The series takes around nine hours.
Skills you'll gain
Prerequisites
Chapters
Function calling, schemas, execution and the foundation of every agent.
The ReAct pattern: reason, act, observe, loop. With stopping conditions, budgets and safety.
Short-term, long-term and episodic memory; task decomposition and planning for complex agents.