A context window is the amount of text and other tokenized information a model can consider in one request. It holds:
the prompt
the instructions
the prior conversation
source files
tool output
the model's own response
Every model has a limit. When the window fills, older content may be removed or summarized, and important details can be lost. Even before the limit, too much unrelated material can make the model overlook the right fact.
Give the assistant current, relevant sources instead of a repository dump. For long tasks, record decisions and progress in a compact handoff. Careful context selection improves accuracy, cost, and speed.
Rewriting in plainer words…
This answer doesn't lend itself to a diagram - it reads best . No credits were charged.
How would you choose which files to give the assistant for a bug fix?Starts from the failing test or stack trace, adds the code and interfaces it touches, and leaves out unrelated modules and stale docs.
What you can say
A context window is the amount of tokenized text a model can take into account in one request.
Everything shares that space: the instructions, my prompt, the earlier conversation, source files, tool output, and the model's own reply.
Every model has a limit, and when the window fills, older content may get dropped or summarized, so details I cared about can disappear.
Even well before the limit, a pile of unrelated material makes it easier for the model to miss the one fact that matters.
So I give the assistant current, relevant files instead of the whole repo, and on long tasks I keep a short handoff note of decisions and progress.
Picking context takes effort, and trimming too hard can hide something the model needed, but relevant context gives more accurate, cheaper and faster results.
Weak answers to avoid
Calls it the model's long-term memoryThe window only holds what is in the current request. Anything not sent, or dropped when the window fills, the model cannot see.
Assumes more context is always betterUnrelated material makes the model overlook the right fact, and it costs more and runs slower. Say you pick current, relevant sources.
Counts only the promptPrior turns, files, tool output and the model's own reply all use the window. Naming them explains why long agent sessions fill up.
Stops at the definitionThe question asks why it matters. Add what happens when the window fills and what you do about it, like a compact handoff on long tasks.
Interview lens
Likely follow-ups, what you can say, and the weak answers to avoid.
