Software is becoming as free and ubiquitous as air. And as it gets easier to make software that is Good Enough™, the need for everyday people to actually touch that software is collapsing.
ChatGPT is replacing Google, and Google is scrambling to become ChatGPT. With agents and MCPs, the next stage of computing won't involve apps or websites at all. Ask your assistant for the best winter coat and it weighs the options and recommends one. The next step is obvious: it asks if you want it. Your payment details are already saved. You say yes, and it shows up at your door.
Soon your phone is just a thin client to a model that knows your bank, your cards, your stores, and above all, knows you. Most of us have already talked to an LLM and felt it caught our meaning faster than most people do. Imagine never scrolling a thousand pairs of shoes again. The model surfaces the one you wanted before you knew it existed.
For shoes, that is close to magic. One pair is interchangeable with the next, and the right one just means the one closest to what you already are.
Travel is not like that. Travel requires taste.
An agent fetches more of what you already are. Read the history, infer the preference, serve the match. Point that at a trip and it breaks in one deep way: it regresses to the mean. It runs on reviews, rankings, and top-ten lists, which are all just averages. The four-and-a-half-star restaurant with twelve thousand reviews is the Amazon best-seller of dinner: safe, popular, and a tourist trap by the time it reads that way. Taste is knowing which signal to ignore. It is reasoned deviation from consensus: the unlisted room, the place that is quietly perfect and badly photographed. The model's great strength, weighing every signal at once, is exactly what drags it to the middle.
And the part that matters most never makes it into the training data at all. Taste is compressed experience. It is the felt difference between a lobby that photographs like a palace and feels like a morgue, and a small room off a side street you would fly back to for the rest of your life. You earn that by being let down in person, hundreds of times, until you can smell it coming. The model has read every review of every hotel on earth and slept in none of them. A bigger model does not fix this. More data only pulls harder toward the average.
Here is what nobody selling you a travel app will admit: this is the same reason the big platforms are mediocre. They are built for everyone, by committee, optimizing the average. And the average, in travel, is mediocrity wearing the costume of consensus.
And the software itself is a relic. The big booking sites still look and feel the way they did fifteen years ago, because they were built then and never really rebuilt. You know it by heart: a wall of twenty near-identical results, a countdown clock inventing urgency, a fake "three people are viewing this room," a price that swells by a third in junk fees on the final screen, and a form that forgets everything the second you hit back. Then the "itinerary" arrives as six confirmation emails you are left to wrangle yourself. We have all just gotten used to it.
So you might expect me to mourn all of this. I don't. The future where you talk to your trip and it takes shape around you is real, and it is going to be wonderful. The drudgery should be automated. Let the machine handle the booking, the syncing, the boarding times, the parts that were always clerical anyway.
What it cannot handle is the part that was never clerical: the judgment of where to send you, and why. As software becomes air, that judgment becomes the last scarce thing. The future of travel is not the end of the travel agent. It is the opposite. It is software finally doing everything it is good at, so that a person can do the one thing it can't.
That is the idea behind Twelve Hours Ahead. Build for the agentic world from the ground up, so your trip is something you can actually talk to, and stand a point of view behind it that no model can generate, because it was earned in rooms instead of scraped from reviews.
We are twelve hours ahead. We are already standing in the future, and we would like to bring you with us.