The context window sets a hard limit on how much a model can work with at one time. A short customer email fits easily, but a long contract, a full email thread or several reference documents can fill the window quickly, especially once tokens from earlier turns of a conversation are counted too.
The common mistake is assuming a bigger context window means a model reads everything equally carefully. Models tend to pay less attention to information buried in the middle of a long input, so a well-designed system still trims and orders what it sends the model rather than pasting in everything available.