Gangsta AI
I Asked 4 AIs to Read My Boarding Pass. Only One Setup Could.
2026-06-30 · 2 min read
I took a photo of my boarding pass and asked a dead-simple question: what time do I land?
Here's the catch that makes it a real test: the arrival time isn't printed on a boarding pass. You get the departure gate, the boarding time, the flight number, the seat — but not the landing time. So answering it requires the AI to connect "flight UA1320, DEN → SNA, boards ~10:40am" to an actual arrival time. That means knowing the flight's scheduled duration, or looking it up live.
The big AIs, one by one: zero answers
I tried it on the major AIs individually, on their own sites, with the same photo. And every single one tapped out:
Four of the most advanced AIs on the planet, and not one gave me a time.
- "That's not on the boarding pass."
- "Check the airline's app for arrival details."
- "I can't determine the arrival time from this image."
The same photo, side by side, with live data: nailed it
Then I dropped the exact same boarding-pass photo into Gangsta AI, where the models run side by side and can pull live data. The elite tier nailed it instantly: lands ~12:50pm at SNA — John Wayne / Orange County. Every one of them landed it. (Pun fully intended.)
Why did the single models fail?
Two reasons, and neither is "AI is dumb":
1. No live data by default. Most single-model chat interfaces don't automatically pull real-time flight data, so they can only read what's literally on the image. The arrival time isn't there, so they can't produce it. 2. Honest hedging. Without the data, the correct behavior is to refuse rather than guess. The models weren't wrong to punt — they just didn't have what they needed, and punting is useless to me.
The actual lesson
The same question failed on one setup and succeeded on another — and you'd never know unless you compared. This is the whole argument for side-by-side AI in one image: the single models weren't broken, they just lacked live grounding and competition. Run the question where the models have real-time data and answer next to each other, and the correct answer falls right out.
It's a tiny example with a big implication: the "AI can't do that" you hit yesterday might just be the wrong setup — not the wrong question.
Try it with your own boarding pass, or any question the big models punt on: Gangsta AI.
Related reading: Why You Should Care About Frontier AI Models · Why Single-Model AI Is a Dead End · Inside an AI Aggregator: Fanning Out to 30+ Models at Once · Benchmarks Lie: How to Actually Evaluate LLMs for Your Use Case