How I used a swarm of AI agents to do something totally mundane — book a vacation rental — and what it accidentally taught me about how people decide.
I wanted to book a condo in Maui. Twelve nights, two people, a birthday. Under $3,500, an ocean view, and — the part that broke everything — a beach you could actually swim at. Soft sand, no rocks.
That last one has no checkbox. No booking site has an "is the beach actually nice" field, and every listing swears its beach is perfect. So a task that sounds like ten minutes of clicking became a real search problem. I pointed a swarm of AI agents at it. It took 249 of them.
Instead of one assistant searching, I ran dozens in parallel — each hunting a different way: Airbnb, owner-direct sites, timeshares, independent beach guides. Then, for every "winner" they turned up, I ran a second wave of skeptic agents whose only job was to try to prove it wrong. Not rank it. Kill it.
Searching gives you lots of plausible answers. Trying to disprove them is what leaves you with a true one.
The requirement that mattered most wasn't one I could have written down. The beach had to be soft and swimmable — no rocks. I only figured that out by being shown a place that was technically "oceanfront" and recoiling, because it fronted a shelf of lava rock.
That's not me being vague. It's how people are. "A good beach" has no definition and no checkbox — you know it when you see it. You often can't list what you want, but you can tell instantly when something's wrong. Every time I rejected an option, I was really teaching the system what I cared about. (The things I could write down — the budget, the dates — I handed over on day one. The work was keeping the machine honest about the stuff that only got sharp once I saw it.)
Which flips the whole thing on its head. The agents weren't a search engine. They were a cheap way to discover what I wanted by bumping into what I didn't — and cheap is the magic word. When re-searching costs a weekend, you commit early and settle. When it costs ten minutes, "I'll know it when I see it" stops being lazy and becomes the smart move.
When the dust settled, nine genuinely-good options were left. Choosing between "beachfront" and "an extra room" and "$500" isn't math — there's no correct formula. So the agents did what agents are good at (enforce the facts: is it real, available, affordable, actually swimmable?) and I did the one thing only I could: pick what I value.
249 agents narrowed a few hundred options down to nine. I picked the final one in about thirty seconds.
We booked a roomy 1,000-square-foot place across the road from the best-rated soft-sand beach in the area, for $2,937. And here's the honest kicker: it wasn't the agents' #1 recommendation. They'd ranked a beachfront splurge first and a cheaper place second — this was #3 on their own shortlist. I overrode them for the one thing they couldn't score: I wanted the extra room and the better beach more than I wanted to step straight onto the sand. It was never the front-runner. It was just the last one standing — plus a human's taste. The whole army of agents cost less than one night of the stay.
You can now aim a small army of tireless, skeptical researchers at the most boring decisions in your life, and have them do the parts people are bad at — being exhaustive, and being suspicious — while you do the part only you can. That's a genuinely new thing, and it's kind of wonderful.