Why AI shouldn't book your Flight

Personal malleable decisions, unreliable agents, and the credit card test.

During a demo for our AI agent, a sales colleague once asked, ‘How does it process refunds? Does it click through the website like a human?’ I grinned and said, ‘Nope, it just calls the refund API.’

Cue the blank stares.

Turns out, most people assume AI works exactly like we do: clicking, scrolling, hesitating, maybe even muttering under its breath when a page takes too long to load. But here’s the truth: the best automation isn’t about mimicking humans, it’s about skipping the theater and going straight to the solution.

Which brings me to today’s AI travel agents.

The Allure (and Illusion) of ‘Agentic AI’

You’ve seen the demos. An AI books a flight, plans a vacation, even snags a dinner reservation, all while browsing the web like a caffeine-fueled assistant. It’s impressive, in the same way watching a robot arm juggle knives is impressive.

But here’s the problem: one hallucination, one misclick, and suddenly you’re flying to Sydney, Nova Scotia, instead of Sydney, Australia.

Sure, in a carefully rehearsed demo, it works flawlessly. But real life? Real life is messy. Websites change. Prices fluctuate. Your 2 AM brain decides "Actually, maybe a beach trip is better than skiing." An AI doesn’t know that. It just knows how to click "confirm" on whatever you told it five minutes ago.

Some Decisions Are Too Personal for AI

I call these "personal malleable decisions". Tasks that aren’t just about execution, but about constant tweaking, gut checks, and last-minute panic. Booking a flight isn’t just pressing "buy." It’s:

Even a human assistant struggles here. How do you explain "reasonable distance" or "I’ll know it when I see it" to an AI? Flexibility isn’t a bug, it’s a feature of being human.

APIs: The Unsung Heroes of Reliability

So what’s the alternative? APIs.

Yes, they sound boring. No, they won’t wow anyone at a conference demo. But they work.

Years ago, we worked with an airline whose website looked like it was built in 1999. Instead of training an AI to navigate its janky dropdown menus, we scraped their data into a clean, reliable API (we had permission!). We even added a "change detector" to alert us if they updated their site.

Was it glamorous? No. But it meant no surprises, no misclicks, no accidental one-way tickets to the middle of nowhere.


The Future of Agents? Maybe. Today’s Reality? Proceed with Caution.

Look, I’m not anti-AI. Someday, agents might handle our messy, ever-changing decisions. But today? They’re like a toddler with your credit card—cute in theory, terrifying in practice.

Until then: save the flashy demos for conferences. For real-world tasks, stick with APIs, scrapers, or tools you control.

Your wallet (and sanity) will thank you.

Next time you see an AI agent "magically" book a flight, ask yourself: Would I trust this with my vacation… or my credit card?


Comments

There are no comments added yet.

Let's hear your thoughts

For my eyes only