There's a lot to explore at the intersection of context engineering and UX!
In my blog post on Context Engineering (link below), I mentioned that I now think about the content of the context whenever I interact with AI apps.
One way this has modified my behavior using chatbots is that I use the "Edit message" feature a lot more.
When I ask a tangential question that won't be relevant going forward, after getting the answer, instead of continuing below, I edit my previous message with my next question.
I do this to keep the context focused and avoid context rot.
My ChatGPT chats now look like shallow trees instead of linear paths.
Using "Edit message" all the time, however, is not a great experience. ChatGPT supports it but wasn't designed around it.
One issue is that the tangential questions in side branches are pretty much lost. I don't want them in the current context, but I may still want to revisit the history later.
One solution is "tree-first" interfaces for chatbots. The chat history is presented as a tree, where you can easily visit and branch off from any node.
I can even imagine more interactive types of context management, where you can mark specific nodes in the conversation tree to keep/remove from the context.
