I’m not a big fan of LLMs as decision-making tools, but they do serve a purpose in my workflow as an organization layer. Specifically, when it comes to travel. The problem is that if you ask Gemini, ChatGPT, or Claude to plan a multi-day trip to Tokyo, you will probably get a generic list of tourist attractions. It might tell you to visit Tsukiji market at noon, suggest a restaurant that closed two years ago, and send you across the city three times in one afternoon. That tracks, since it’s just basing its recommendations on the collective knowledge of the internet — both old and new.

AI doesn’t really know your preferences, how much walking you can handle, or whether visiting a particular spot actually makes sense given everything else on your itinerary. I figured this out earlier this year while planning a three-week trip to Japan.