Agent Engineering
By Wanderson Leandro de Oliveira
An AI agent is not a prompt trick — it is a piece of software with a loop, a state, a set of tools it can call, and criteria for when to stop. This course builds that software from the ground up in plain Python, with no agent framework standing between you and the mechanics: the perceive-reason-act-observe cycle, native function calling across providers, planning with ReAct, short-term memory and context window management, long-term memory backed by vector search, and retrieval-augmented generation wired in as just another tool. Three hands-on lessons evolve a single command-line agent from a bare loop into a tool-using, retrieval-augmented assistant, so by the end you understand every component well enough to know exactly what a framework like LangChain or CrewAI is automating for you in the next course of the track.
Course content
The Agent Loop
- 🔒 The perceive-reason-act-observe cycle text
- 🔒 Termination, budgets and loop detection text
- 🔒 Hands-on: building an agent loop without a framework text
Tools and Function Calling
- 🔒 Designing reliable tool contracts text
- 🔒 Native function calling across providers text
- 🔒 Hands-on: building a tool-calling agent text
Planning and ReAct
- 🔒 Task decomposition and planner-executor architecture text
- 🔒 The ReAct pattern text
- 🔒 Reflection and self-critique text
Agent State and Short-Term Memory
- 🔒 Modeling agent state text
- 🔒 Managing the context window text
- 🔒 Context compression strategies text
Long-Term Memory and Vector Search
- 🔒 Semantic, episodic and procedural memory text
- 🔒 Embeddings and vector search for agents text
- 🔒 Write policies, forgetting and privacy text
RAG for Agents
- 🔒 Agentic RAG patterns text
- 🔒 Grounding, citations and faithfulness text
- 🔒 Hands-on: adding retrieval as an agent tool text