Context Engineering

Overview

Your agent works in the demo, then gets worse on real work: it forgets instructions halfway, picks the wrong tool, or drowns in documents. That is almost always a context problem. At each step, the model only knows what is in its context window, so the performance and quality of an agent depend on what it receives there. In this course you will learn the art of Context Engineering: giving the agent the right information, at the right time, in the right amount. You will also get familiar with common methods such as MCP, skills and RAG, and how to build them.

What you'll cover

• ▸What makes up context, such as the system prompt, instruction files, tool results
• ▸Why context is a finite resource and how to avoid quality drops on long context
• ▸When to use the context window and when to use external storage
• ▸How to connect agents to tools and data using MCP, the problem it solves, its architecture and primitives
• ▸What a skill is, how to write a good one and the benefits of progressive disclosure
• ▸What subagents are, and how they help with isolating context and reducing main context bloat
• ▸What RAG is, and the ingredients that make it tick: chunking, embeddings and similarity search

What you'll be able to do

• ✓Choose the right context engineering approach for a task
• ✓Explain how context and memory shape agent behavior, and find failures caused by too little, too much, or the wrong context
• ✓Manage context for long running agentic tasks
• ✓Build and use MCP servers
• ✓Write skills that use progressive disclosure and code execution
• ✓Design and build a RAG pipeline