用 Muse、Instinct、Dot 全程规划一次旅行

Manosai @manosaie原推发布于 2026.10.27

在 X 上打开原推

I decided to be the dream personal agent user and plan an entire trip end-to-end using AI. I am the unrealistic unicorn user, representing 0.001% of American consumers…context-maxxing the hell out my agents (mostly Muse, Instinct, Dot, and a little GrokBot) and doing crazy things like traveling more than once a year. Here are my raw thoughts. pros: - doing trip planning / research with agents is awesome bc of the time saved from fan-out style search + information retrieval. This technically existed with chat-based assistants but two things meaningfully improved: 1) browser use/voice/long horizon has meant agents can go farther in information retrieval + plan execution than before 2) personalization is getting better bc these agents have more context + memory about you than strict chat-based knowledge gathering. after a little bit, the agents also got the hang of my preferences and feedback and I did feel like the trip “was coming together” in a way that reflected convergence towards a goal - Muse had the best UX around trip planning in a way that felt natural. I loved having a side chat around this trip that I could keep coming back to, it was the most natural form of recall. I don’t want everything in my life to be a project, nor do I want it feel like a goal. These containers might be the correct UX level primitives to name, but they just feel like things I need to remember and go back to easily — and I think Muse takes the cake here - Voice calls with Dot for random brainstorming was dope - the day before the flight, I needed to change it to a later departure time due to a last minute emergency. I got one agent to do the math on how much canceling altogether would cost me, and two others to start chatting with airline support to figure out alternatives. I tried to do this myself but was on hold over the phone for over an hour, and never got anywhere. Being able to delegate in time-sensitive situations is an incredible life hack. Instinct got it done and rebooked us on another flight, trip saved the cons: - verification is hard, and you don’t know where a human or an agent messed up. In-person check-in was difficult bc apparently somehow our baby’s ticket was dropped altogether on the new flight. No one at the checkin counter understood how it happened, I didn’t have the heart to tell them how involved my agents were in coordinating the change. We also had issues on the way back because apparently our baby was now double booked 🤷🏽‍♂️. Hard to decipher since I don’t know what my agent discussed with support — perhaps we need full receipts or a way to audit more accurately, and give feedback to close the loop. - closing the loop on trip planning is also hard. So much of the eval on whether the trip went well or not is based on in person context. My agents won’t get any of the feedback unless I explicitly tell them, or if they infer from my in-trip activity. Both are not great. - in-trip use basically sucks. It’s either too slow or not convenient enough to rely on agents when you are in the moment and need to make decisions. This feels like a feature though, humans are still way better at navigating real life and figuring shit out. that’s the point of every trip anyways, you can plan all you want but ultimately, you have to go with the flow. So I’m okay with this, bc it meant I stayed off my phone in the right moments. - however when I did want something real-time, Instinct in its text message form was damn convenient. Something about it still just feels far more reachable than my other agenda. My hypothesis is text messages are already a “real time context area” so having an agent live in that space made it more too of mind to reach for That’s all for now, fun use case