Context Window.
How much a model can hold in its head at once, measured in tokens. If you go beyond this magical limit, the model starts forgetting what you told it just before. However, it won't tell you when it's reached that point. As technology has improved context windows have grown, but it's worth noting they're readily eaten up, especially in a chat with a lot of files (especially token-hungry ones), and also in chats with a lot of turns.
In the wild
“Irrelevant information does not simply sit harmlessly in the context window; it can distract the agent and degrade its decisions”
Seen here: Forbes, 2026
For AI users
It's easy to think that once you've told AI something in your chat, it doesn't forget. But even if it doesn't tell you it may well have forgotten some context that is critical to you. There are some methods of use that optimise context and others that drain it, like uploading a lot of colourful PDFs (the colours, fonts and so on all add to the depth of the context but they might not be important). Tips for optimising context include using markdown files and specific project context upload features, and also reducing the number of turns in a chat through more effective prompting, including using voice. Or using other methods entirely, like knowledge graphs.
For business leaders
Make sure the team understand these foundational concepts for how to operate AI well, and that they have a way of understanding how to improve as the technology changes or our knowledge improves. This training can't therefore be static. It needs to be refreshed with the latest tips regularly, and using department-specific AI champions can be a way to get more up to date training to our teams more regularly. As context contributes to token use, it makes financial sense to optimise context, to improve your token to impact ratio and hopefully therefore your ROI. When running into context window issues (and risking hallucinations) it's worth knowing other AI based methods or steps might be more appropriate.