The context window is the boundary around the AI’s working memory. Everything the model holds at once (instructions, the conversation, file content, tool output) has to fit inside, and that limit is measured in tokens.
Picture an employee in a lecture series. Drop them into seven lectures in one day and test them at the end, and they’d remember the main ideas but miss the detail from hour three. Overload hides for a while: they sound confident on the easy questions, and the misses appear when an answer depends on an earlier detail.
Why you care
In a long session, the AI may forget earlier details or mix instructions. The model didn’t get worse; the session got crowded. That’s why long work gets split into phases, why subagents research narrow questions in their own smaller context and report back, and why tools use compaction to summarize when the window fills (helpful, but lossy). Skip memorizing exact numbers; they change by model, product, and plan. Keep sessions focused, save durable context into files, and wrap up before the room gets crowded.