Prompt Caching: Notice What Changes Between Requests
This Short looks at a detail that is easy to miss when an AI conversation feels like one continuous workspace: requests can reuse some input, and changes to the setup can affect that reuse.
In the video, I discuss pauses, model switches, and fast mode as things to pay attention to. I also distinguish editing instruction files from those other changes in the Claude workflow being discussed.
Those are observations about a particular product setup, not a rule that applies identically to every AI service. Cache behavior can depend on the model, client, cache lifetime, and how instructions are assembled.
The experiment I would keep is simple: change one thing at a time. If you switch models, note that separately from taking a long break or changing modes. Where the tool exposes cache usage, compare those readings.
That gives you something more useful than guessing from the apparent speed of an answer: a record of what changed and how the service actually processed the request.