I prefer an AI that finds the right context to one that drags my entire history into every answer. More memory also means more old decisions available to show up in the wrong place. I want to know why a piece of information belongs to the current request, especially when the subject changes.
Anthropic's April 2026 research on personal guidance puts this product choice on the table. From an agent, I want initiative to search and judgment about what to use. The task does not need to become a package with everything that happened before. The interface needs to let the subject actually change.
A common case in engineering: an agent picks up a deadline constraint given for one project and carries it into the next one. Without its original situation, that constraint looks like a permanent preference of mine. The next answer rests on a premise I do not recognize, and I have to find out where it came from.
The objective defines what enters the context. If the answer is in a relevant, authorized reference, I expect the agent to read it. Access to a folder of files is no reason to use all of them, and old text should not fill a scope decision I have not made. When an answer uses another conversation, I want that link visible and simple to correct. A large history I cannot revise leaves me arguing with old conclusions while I try to ask a new question.
When I assess this kind of assistant, I will demand control over the references it uses. I will look for ways to limit the context and correct an association, alongside what it remembers.