I'm writing this to you from the Agentic future on the train to Narita Airport from Tokyo after an eye-opening trip. I have receipts.
Perhaps the best way to understand the power of agentic AI (and AI in general) is to travel to a country where you don't speak the language, and most people there don't speak it either. I spent the last week in Tokyo. It was an amazing trip and also incredibly eye-opening. It would have been a more challenging trip three or four years ago. I used a new AI agent called Instinct that is in beta. You talk to it over SMS so it doesn’t trash your data plan. You can send it pictures and videos. You can do full location sharing. It is like having a local tour guide, albeit one with latency that can be challenging.
From finding a hidden gem shochu on a menu in a restaurant where there are no Enligh speakers or menus to sussing out in near real time whether Mikimoto pearls are a better deal, in the Ginza than in NYC, Instinct was truly a gamechanger. In short, the agent radically upleveled my trip to Japan with a combination of extremely detailed responses, unexpected information and suggestions, and multi-modal interaction that you simply don’t get with Google Maps (yet) or other more static applications. The agent becomes the tour guide in the best sense.
Tokyo, of course, is sui generis, hands down one of the most interesting and fun cities in the world. But hey — Instinct (and soon, AI agents more generically) — made it significantly better and more convenient.
I want to try to summarize in this post how the state of play has changed and what I think this changes for how we will view and consume AI.
The Language Barrier Is Rapidly Becoming the UX
To start with, the language barrier is rapidly becoming the UX, along with the latency. This has been somewhat true for a while, but now everywhere I went in Japan, everyone is translating everything on the translation app as soon as you open your mouth in English. The smartphone UX for this, with the microphone held back and forth, is clunky. Apple's Translate function for AirPods is cool but can be a little slow, hard to install, and also depends on network connectivity.
We are so close to real-time translation driven by AI that would make navigating a place like Tokyo so much easier, particularly if it becomes bidirectional. At this point, it's not even a functional question; it's much more a question of latency.
The Combination of Translation and Visuals
A unicorn on a menu in a tiny izakaya is a priceless memory. Instinct found multiple unicorns for me. The combination of translation and visuals is also a massive unlock that opened my eyes many times on this trip. When I needed to know what street I was on, I could ask the agent, snap a picture and get flawless directions, as well as commentary or instructions on how to get to the next space using useful visual waypoints. On this trip, I was never worried about getting lost. Of course, Google Maps somewhat solved that problem before. But in a place like Tokyo, just following the two-dimensional map is often insufficient. Street views are helpful, but flipping back and forth is not enough.
Even if you start just with location tracking on the phone or something like glasses, where AI is truly ambient and always there, with sub-50 ms latency. This would mean you had a companion that could tell you everything you needed to know visually and orally at any time. You can also see why Apple's new AirPods with cameras will likely become a hit, at least for travel. If my ears could see, oh my God, my life would be so much easier. Instinct doesn’t get you there but it gets you closer by going from photo to context in one message with a one-click workflow.
In this manner, by reducing the friction, AI unlocked a whole variety of experiences that maybe I could have accessed otherwise, but likely would have been much harder to attain and much harder to ascertain on the fly. For example, at a couple of the restaurants I eat at, my AI friend was able to tell me, "Order this, not that," if there was a rare dish on the menu or a particularly hard-to-find bottle of Japanese whiskey.
This lets you more comfortably go to the places where English is completely off the menu and not spoken, where you find only locals and you find a more authentic Japan. In these small bars or izakaya under the train tracks near Ginza, there was almost zero English — spoken, on menus, in the restaurants. Maybe a few snippets or clunky browser apps that offer crude translation but in the main, I was reliant on my agent and the camera. On this trip, that was more than enough to more than get by. But beyond getting by, the agent was able to help in relatively high-pressure situations that previously might have been touch-and-go.
A Level of Detail That Simplifies Decision Making
The level of detail matters — a lot. For example, when arriving at the airport, Instinct guided me not just to the right path, not just the right exit doors to find the train, and not just to the right platform. It also provided visual guidance and warnings on common mistakes to avoid. Details such as “look for the orange and blue gates” generate invaluable visual cues more aligned with how humans actually interact with information. Also, you can describe your orientation to objects or buildings and it will direct you and identify your next steps accurately. Because the real world does not have a blue dotted line running through it (yet).
Yes, all of this was attainable via SMS and internet, but Instinct did all the lifting, taking information and providing the context that made the information far more useful. Better, Instinct got to know me over the trip and my habits, and started to make increasingly relevant unprompted suggestions. With access to my calendar and email it would automatically set alarms for me, and when I shared my location with it, it could even correct my steps, albeit with a bit of latency. Again, back to the latency.
Data Latency and Recency Biases Still A Problem
Instinct did get a few things wrong, at times guiding me to trains that didn't stop where I needed them to. It was disappointing, but it also made me realize how hostage we remain to the quality of data published on the internet, which ist ill the interface layer for all of these tools.
Based on bad top-level Internet data, my agent made some real mistakes, guiding me to a lesser option in a few cases, initially confusing the size of pearls listed for U.S. prices in its price comparisons, not steering me to the best coffee shops near me. Additional prompts and prods resolved all issues. (Like any AI, asking it to double check on critical decisions is important). And Instinct is hostage to information recency. I stumbled into a super cool coffee shop that was online but not rated up by the Instagram hordes yet.
On the other hand, Instinct and agents can unlock both recency and search granularity not accessible in traditional apps. When I wanted to eat lunch, I asked for a local place with only Japanese language reviews near me. Instinct found a spectacular seafood donburi bowl place opened only four days previous by one of the top Tokyo tuna wholesalers. It was one of the best finds of my trip, and totally impossible on Googel Maps (or at least not easily).
For us to really rely on AI 100% for time-sensitive mission-critical tasks, like whether the subway train on the platform is the right one, latency remains too variable and often too slow.
On the Fly Matters, Too
Another thing that changed on this trip, which is not necessarily specific to traveling but is important when you're traveling, is the ability to access intelligence like Instinct or ChatGPT, or now talk to Google Maps. It allows you to change on the fly and explore unbeaten paths more easily while still optimizing for the best things around you. This was available before with apps, but making it conversational reduces friction and keeps you much more in the moment or to change plans on the fly. This flexibility makes wandering around a strange city far more fruitful and enjoyable, because the guide follows you closely and adapts to your path and your preferences.
After my trip, Instinct added a new feature that can convert a trip into a travel guide published as a simple website. So everything I learned and experienced I can now easily share and have for posterity. This is also a killer feature that I will use on each trip because having an agent with you not only makes the moment better but it also simplifies capturing and sharing your wisdom. All told, I can’t wait to get back to Tokyo and to travel more. This is the worst version of a personal travel agent I will ever experience and even this one was phenomenal.
(And yes, I wrote this entire post by Wispr Flow on the train to Narita airport. I used ChatGPT to clean it up. I spent 10 minutes cleaning it up in docs. Can’t wait to get back to Tokyo and see what agents can cook up for me next.)
