I Let an AI Browse the Web for Me While Planning New Year’s Eve in Bangkok
Most travel planning tools help you discover places. Very few help you deal with reality — prices, availability, booking rules, and last-minute constraints.
When I tested Perplexity for the first time, what stood out wasn’t the answers themselves, but the way it arrived at them. This wasn’t just an AI responding to prompts. It was an agent actively browsing the web on my behalf — opening pages, scrolling, cross-checking details — while I followed along through screenshots of what it was seeing and interpreting.
I put it to work on a deliberately difficult travel scenario: New Year’s Eve in the capital of Thailand with many rooftops and cool places.
Starting With Realistic Constraints, Not Inspiration
I didn’t ask for “the best rooftop bars.” I asked for something I could actually book.
The brief was specific:
- Rooftop bars in Bangkok on New Year’s Eve
- Maximum budget of €400 per person
- Central districts only
- No ultra-luxury venues
These kinds of constraints usually force you into hours of manual checking. That’s where I wanted to see how far the agent could go on its own.
Watching the AI Do the Browsing
Instead of opening dozens of tabs, the AI agent navigates event pages and booking platforms on your behalf, scrolling through New Year’s Eve listings and reading the same details a human would.
Instead of instantly producing a list, Perplexity began navigating the web.
It opened venue websites, scrolled through New Year’s Eve event pages, and jumped between booking platforms. I could see what it saw through screenshots, and it was clear that it wasn’t skimming headlines — it was reading the details most people end up hunting for.
This was the key shift. Rather than asking an AI to summarize information, I was watching it look for information.
Narrowing Down Rooftop Bars That Were Actually Viable
As it browsed, the agent filtered out venues that didn’t fit the brief.
High-end rooftops with €400–€800 packages disappeared quickly. What remained was a list of around ten rooftop bars in central Bangkok that were popular, realistically priced, and positioned somewhere between casual and luxury.
At this point, it already felt closer to a shortlisting process than a recommendation list.
Checking Availability in Real Time
After reviewing multiple pages, the agent condenses the findings into a structured overview. Prices are cross-checked against the budget, availability is verified, and package details like inclusions and time windows are pulled out explicitly.
One of the most useful steps came next.
Perplexity didn’t assume availability. It checked it.
For each shortlisted venue, it looked at whether New Year’s Eve tickets were:
- Still available
- Already sold out
- Limited to certain time slots or package types
Anyone who has planned a holiday event knows how misleading outdated travel articles can be. This step alone eliminated many false options.
Turning Pricing Into Something Comparable
Instead of throwing out a single number, the agent summarized what each ticket actually included.
That meant breaking down things like:
- Drinks included or not — and whether they were free-flow
- Food included, optional, or not available
- How long the package lasted
- What kind of event structure to expect
This turned what is usually a confusing comparison into something clear enough to evaluate at a glance.
Understanding How the Agent Works (A Quick Detour)
To make sense of why this felt different, I later tested the same agent on a non-travel task — navigating an inbox. The point wasn’t email productivity, but seeing how it moved through an interface, identified states, and acted on what it saw.
That helped clarify what was happening during the travel research. The AI wasn’t “guessing.” It was observing, navigating, and interpreting live environments — the same way a human would, just faster and more systematically.
Back to Travel: A More Honest Kind of Research
When I asked a follow-up about LGBT-specific rooftop events, the agent didn’t invent options that don’t exist. It explained that Bangkok doesn’t have rooftop bars dedicated solely to LGBT New Year’s Eve events, but pointed to inclusive venues and celebrations that are known to attract diverse crowds.
That kind of grounded response reinforced the feeling that this was research, not marketing copy.
What This Changes for Travel Planning
This doesn’t replace local knowledge or personal taste. It won’t tell you which rooftop has the best energy at midnight.
But it handles the tedious, error-prone part of travel planning — checking, verifying, comparing — with a level of patience most people don’t have.
For event-driven travel, especially around peak dates, that’s a meaningful shift.
So, How Was It?
This was my first real experience with an AI agent that didn’t just respond, but acted.
Watching it browse the web on my behalf changed how I think about travel research. Less searching. Fewer assumptions. More clarity.
This won’t change how I approach search — at least not yet. But as a first encounter with a different kind of online workflow, it feels like the beginning of something shifting. The speed is still a limitation, but there’s a trade-off: the process is visible.
I can follow what the agent is doing, see where information comes from, and understand how conclusions are formed. That transparency, more than the output itself, is where the real potential starts to show.
And once you’ve seen that happen in real time, it’s hard to go back to asking an AI to simply “list recommendations.”
And which travel app do you use?
