Vibe Coding an OTA, Thoughts on AI:

We appear to have entered a new era for code generation. Since December I’ve frequently seen some impressive vibe coded solutions on twitter, so I wanted to take a crack at it myself. When traveling for work on long haul flights I try to find lie-flat seats, but despite Google Flights having an indicator for lie flat, you can’t actually search by it. So I decided to vibe code an OTA focused exclusively on lie flat seats: SomniSeat.

What surprised me wasn’t that I could build a lie-flat-only OTA in a day, it’s that nothing structurally prevented it from existing before. The constraints weren’t product vision or capital, they were developer time, integration friction, and iteration speed. Those constraints are collapsing fast.

All of this took about one day’s work utilizing Claude Code, GitHub, Vercel, and the Amadeus self-service API for data. I have minimal coding experience beyond PHP back in the day and it spit out an impressively good copy of an OTA including some cool pages like Inspiration (pulling recently searched travel), AI Search (utilizing OpenAI API and facilitating search of the underlying data), and Lie-Flat Routes (pulling cached data from searches to demonstrate where and who flies lie-flat business class seats.

I will caveat that I haven’t hooked up the actual booking mechanism (either through Amadeus or sourcing a meta-esque solution through KAYAK or direct with airlines) mostly because I don’t want to deal with the hurdle of contracts and legal setup since this is more playing around. Additionally, Amadeus is pretty ass-backwards from a technical perspective and the ability to move from limited test API data to production has been having technical issues for awhile and even the test API goes down sometimes so it’s a bit frustrating. Hence the actual flight results are limited right now, it’s probably 20% of actual flight availability and mostly close in with cached data but I’ll fix it as soon as Amadeus does.

More thoughts: