Core concepts: What is the context window

The single most important concept in getting useful work out of AI, and why most people get it wrong.

Every AI conversation you’ve ever had has happened inside a context window. It’s the model’s short-term memory: the sum of what you’ve told it, what it’s told you, the system instructions it’s running under, any files or documents you’ve attached, and any tools it’s called. Once that window fills up, things start going wrong — answers get vaguer, details get lost, behaviour gets weirder. Understanding this one concept changes how you use AI forever. Aimed at anyone who has used Claude, ChatGPT, or any AI tool and hit the wall where it suddenly feels stupid. That’s almost always a context problem, not a model problem. No technical background required — the concept is simple once it’s explained clearly. By the end of the 90 minutes you will: • Know exactly what a context window is, what’s in it, and why models degrade when it fills. • Know why "prompt engineering" is becoming "context engineering" and what the difference actually is. • Have strategies for working with a limited context window: compression, retrieval, sub-agents, knowing when to start fresh. • Be able to diagnose a stuck or confused AI session and fix it, rather than giving up and starting a new chat. How the session runs: 25 minutes on the concept itself — what tokens are, how context is assembled, why it fills up. Then we open a real Claude Code or Claude session and visualise its context piece by piece. Back half is practical: the five techniques that make long sessions not fall apart, tried live on whatever you bring. Bring your laptop if you want to try the techniques on your own sessions. Works either way — you can just watch.