Every now and then, you hand a blank canvas to a group of assistants and realize they’ve all been reading the same travel guide. That’s the odd feeling I got when I recently tested a collection of AI agents on a very personal, very open-ended assignment. My partner and I are thinking about exploring China late next year, a two-week-long adventure that should feel both historically rich and energetically modern. So I sent a long, detailed paragraph to six different AI assistants, explaining that we wanted a first-draft itinerary with flight, hotel, nightlife, and food recommendations. I asked them to chart a course that blends historical moments with big-city vibes. I also made it clear that I can’t stand constantly changing locations, so they should limit us to four different hotels maximum. The idea was to see how these tools handle a broad, human request—not a simple prompt like “write a poem” or “summarize an article,” but an actual planning task with constraints, preferences, and a bit of personality. The results were convenient, polished, and almost suspiciously unanimous. Every single agent chose the same route through China: Beijing, Xi’an, Chengdu, Shanghai. No alternate versions, no “have you considered Guilin or Yunnan?” No one asked whether we wanted to avoid crowds or spend more time in nature. They all landed on the same safe, sensible, historical-meets-modern loop. It was a little like watching a group of travel agents who had all been trained on the same guidebook, and it made me wonder if we’re entering a new era of what we might call the Lonely Planet effect—except this time, the guidebook is a neural network, and the herd mentality is written into the algorithm. Instead of individual discovery, AI agents are nudging entire groups of travelers toward the same locations, shops, and restaurants.
On the itinerary test, the scores were close but the details differed in ways that mattered. Muse took the top spot with a perfect 5/5. Its draft was extremely easy to read, with visual tables and straightforward tips. It suggested flying out of San Francisco the week after Golden Week, on a nonstop flight to Beijing, and then returning home nonstop from Shanghai. Four hotels, no backtracking. Imperial history first, Sichuan food and pandas in the middle, Shanghai nightlife to finish. It was the kind of itinerary you could actually show a partner without having to explain it. Dots came in second with a 4/5. Its draft was strong but excessive, clocking in at 24 pages—impressive, but more than anyone needs for a first pass. Its core recommendation was to fly into Beijing, take high-speed trains to Xi’an and Chengdu, then fly to Shanghai and home. Dots even highlighted that you’d get imperial history, the Terracotta Army, Sichuan food, and a big-city finale, with no overnight train and no extra hotel moves. Grok Bot also earned a 4/5. Its output was easy to read and full of realistically helpful tips. It proposed a first-draft 13-night trip for two: Beijing, Xi’an, Chengdu, and Shanghai, departing SFO on Oct 13 and returning Oct 27, after Golden Week. That felt like a sensible calendar. Gemini Spark got a 3/5; the itinerary was fine, but it lacked the readability and finer details of the higher-ranked agents. Its title was a mouthful: “A First-Time Traveler’s Route Blending Imperial History, Sichuan Gastronomy, Vibrant Nightlife & 4-Hotel Hub Comfort.” Instinct also got a 3/5. The trip was decent and helpful, but I preferred the layouts from other agents. Instinct’s one-line summary was neat: “Beijing in, Shanghai home, with Xi’an and Chengdu between: four bases, no backtracking, history up front and the food and bars getting better every stop.” Claude Duet rounded out the middle with a 3/5. Despite not using images or visual elements, its itinerary was solid: fly into Beijing on Saturday, Oct 9, 2027, then travel Beijing → Xi’an → Chengdu → Shanghai, and fly home from Shanghai on Saturday, Oct 23. That was 13 nights in four hotels, two bullet trains, and one short flight. Solid, but not soul-stirring.
After the China exercise, I had trouble sleeping. The Vietnamese coffee I’d had that evening probably didn’t help, but neither did the strange feeling that a dozen invisible assistants had all been channeling the same concierge. I started thinking about what happens when you give these tools a broad, open-ended task—especially when you’re not even sure what you need. Sometimes the moments I need the most help are the ones where I don’t know where to begin. At 2 a.m., the usual advantages of a human assistant disappear: I would never message a real person that late. But an AI agent doesn’t mind. So I ran a second test, a kind of insomnia check. I asked the six agents for help with the vague, messy feeling of having too much on my mind. The results were telling. Grok Bot was the clear winner with 5/5. Its response made sense to me, and if the scenario were real, I would likely follow through on it. It even offered to set a reminder for later in the morning to check on my mom’s leg, just to get that worry off my mind. Dots was close behind with a 4/5. Both Dots and Gemini Spark suggested a mental download, but Dots’s phrasing felt more supportive and direct. “Send me the three things taking up space in your head,” it said. “Messy is fine, and I’ll turn them into a short checklist with one clear first step. Then you can put the phone down.” That felt like empathy, or at least a simulation of it. Instinct also got a 4/5. Unlike some of its other answers, this suggestion felt reasonable: “Here’s one you can do from bed: skim the China draft and tell me the one thing you’d change. I’ll have a v2 ready when you wake up.” Muse earned a 4/5 as well, with another strong response. It offered to cross-check my calendar and self-sent email reminders for conflicts, only surfacing anything at a sane hour unless something was urgent. It even apologized for the check it should have done properly yesterday. Gemini Spark, though, dropped to a 2/5. The advice was fine, but the comment about my empty morning calendar was annoying. Insomnia isn’t just tied to what’s on Google Calendar. Claude Duet also got a 2/5, giving one of its weakest answers. It wanted to close out yesterday’s two loose ends—a delivered work pitch and a lunch cart order covered by DragonEats. I was not lying awake worrying about lunch. The difference between the four stars and the two flops was simple: the best agents responded with warmth and a clear next step, while the worst were technically correct but emotionally tone-deaf.
After both rounds, I sorted the agents into categories that reflect how I’d actually use them. Best Overall goes to OpenAI Dots. It is still behind an expensive paywall, so I would advise most people to wait until OpenAI launches it for free, or at least makes it available on the $20-a-month plan. Even so, Dots performed the best across both tests. In the month since its release, it already feels more polished than many more established products. I would prefer a separate app over the current ChatGPT integration, though the smartphone widget is a decent compromise. Dots for the win. Most Accessible is Meta Muse. Muse is for the normies, and that’s fine. It did well in tests, coming a close second on the scorecard, and it was the easiest to set up. Plus, it’s free. But I still have privacy concerns. Despite Meta’s assurances that it responsibly processes user information in secure virtual machines, the company’s past mishandling of user data makes me hesitate. Best Interface goes to Anthropic Claude Duet. I tested this agent just one day after it launched as a research preview, so a few rough edges were expected. Yet its dedicated app was extremely user-friendly, both to navigate and to use for follow-up tasks. I could easily see users obsessed with productivity and streamlined software experiences latching onto Claude Duet in the next few months. You currently have to sign up for Anthropic’s waiting list, and the tool sits behind a $20-a-month paywall, but the interface is genuinely good. Most Annoying, for me, was Instinct. You might like its sparse design and the fact that it lives inside your messenger. I did not. Other WIRED writers have lauded Instinct, even saying it was so good they didn’t care what happened to their privacy. But the texting-only aspect felt like a deal-breaker. I left the experience feeling too removed from what my agent was doing on my behalf, and I was consistently annoyed by its emoji reactions. When I told the agent it had missed a scheduling conflict, it responded with a grimace face. That wasn’t helpful; it was just grating.
What did I learn from all of this? AI agents are extraordinary tools, but they are not humans. The China test showed how easily they converge on the same consensus itinerary. The insomnia test showed how quickly they can forget that the person on the other end of the screen is a person. The best agents—Grok Bot with its caring reminder about a family member, Dots with its gentle invitation to offload messy thoughts, Muse with its careful timing, Instinct with its small, actionable nudge—understood something that the worst agents missed: a request for help is often a request for comfort, not just information. When you’re lying awake at 3 a.m., you don’t need a lecture about your calendar or a reminder about lunch. You need someone to say, “I’ve got this. Put the phone down.” The same applies to travel planning. The itinerary is not the trip. The memories are made in the unexpected moments, in the small noodle shop that isn’t in any guidebook, in the stray conversation with a stranger, in the wrong turn that leads to a hidden courtyard. If every traveler uses the same AI agents, the same prompts, and the same accepted routes, we risk erasing precisely those moments of serendipity. The Lonely Planet effect is about to become much more powerful. When a guidebook tells you to visit a place, it fills up with tourists. When an AI tells you to visit the same place, it fills up faster, because everyone is getting the same answer at the same time. That doesn’t mean we should abandon these tools. It means we should treat them as starting points, not scripts.
So where does this leave me? I’m not ready to hand over my travel plans—or my 2 a.m. anxieties—to a machine. But I’m also not going to pretend that these agents aren’t useful. For a first draft, for a gentle nudge, for the lonely moment when you need someone to hold a thought so you can sleep, they are surprisingly capable. If I had to choose one today, I would pick Dots for its overall performance, but I’d tell anyone on a budget to start with Muse. If you care about interface and follow-up workflow, Claude Duet is worth the waitlist. And if you don’t mind texting and a few quirks, Instinct might work for you—though it didn’t work for me. As for the China trip, I might actually use that unanimous route as a starting point. Beijing’s imperial grandeur, Xi’an’s Terracotta Army, Chengdu’s pandas and Sichuan food, Shanghai’s nightlife: it’s a good route, even a classic one. Four hotels, no backtracking, high-speed trains between cities, a nonstop flight home. But I wouldn’t stop there. I’d ask the agents for the version they didn’t write—the route that leaves room for accident and discovery, the one that doesn’t just connect dots but breaks them. I’d also remember that the best travel advisor, like the best late-night companion, doesn’t need AI to know that sometimes the kindest thing you can do is take a few thoughts off someone’s hands and tell them to rest. The machines are getting better at that, but they still have a long way to go. Until then, I’ll keep one eye on the itinerary and the other on the small moments that no algorithm can predict.
