Over the past several years, I've worked in two businesses that leverage AI in one form or another to solve customer challenges. I'm keenly interested in the debate over the technology's utility, impact, and future, and I'm more informed than most on the topic. Some people that think AI is a technological panacea that'll usher the species into a lasting era of bliss. Others believe AI will be cause the end of humanity. And at the moment, it's wildly overhyped by people who have an vested interest in the technology's success, which is a real problem.
The truth, as is so often the case, lies in the middle. AI is a tool. It can be very useful. It can also lead to some very mediocre if not downright stupid outcomes.
Case in point, the rest of this blog, in which I'll illustrate the useful and the mid over the course of one recent interaction with Claude.
I was curious about whether I could use AI to help plan a road trip. In this case, a month-long tour of the country centered around seeing musicians I like in smallish venues in smallish cities/towns. I started my exploration by offering Claude the following prompt:
"i am a big fan of singer-songerwriter musical acts, like rhett miller, andrew duhon, jeff tweedy, jay farrar, cat power, carbon leaf, brandi carlile, tyler childers, noah kahan, waxahatchee, avett brothers, haim, etc. i'd like to take the month of november and tour the united states and listen to live music."
please create an itinerary for my that starts and ends in northern virginia that allows me to see artists live in mid-sized or small cities across america. make sure i spend at least two nights in each location and get to see at least one live performance. the venues should be small to mid-sized, as well."
It took me four or five iterations and refinements to get to an itinerary that I liked. After I'd reached that point, I asked Claude to, "synopsize this entire series of prompts and responses as a blog post, using the tone of voice of gheorghe: the blog (https://gheorghe77.blogspot.com/). make the itinerary the last thing in the post."
What follows is Claude/Gheorghe's response - my reactions will follow in italics:
So I've got this problem, and the problem is that I like exactly one genre of music with the burning intensity of a thousand suns, and that genre is "sad and/or earnest people with acoustic guitars," and the sub-problem is that the specific practitioners of this genre I love most — your Rhett Millers, your Tyler Childerses, your Waxahatchees, your Carbon Leafs (Leaves? the Carbon Leaf discography?) — do not, as a rule, play the Xfinity Center. They play rooms that hold four hundred people and smell vaguely like spilled Yuengling. Which is exactly how I like it, but it does mean that "catching a show" requires actual research, unlike, say, Taylor Swift, whose tour schedule you could find by asking a golden retriever.
Okay, so this is...decent. Not sure why Claude thinks I only like one genre of music. The voice is an amalgamation of Gheorghies. I definitely spy some OBX Daveish stylings in the mix.
Naturally, since I am both lazy and a person living in the Year of Our Lord 2026, I asked an AI to plan me a month-long, coast-spanning music pilgrimage instead of doing it myself like some kind of caveman with a road atlas. What followed was four entire drafts of a road trip, escalating in ambition and geographic absurdity, and I feel like you should get to enjoy the ride too, because honestly it was a journey (pun very much intended, I regret nothing).
Again, solid. A little try-hard, if I'm being honest. I hope we don't sound like that.
**Round one: November, start/end in Northern Virginia.** Reasonable ask, I thought. Big loop through Richmond, Asheville, Athens, Nashville, New Orleans, Austin, Memphis, Louisville, Knoxville, back home. Charming. Comprehensive. Almost entirely make-believe, as it turned out, because — and this is a very "living in the future" kind of problem to have — none of my guys had actually announced November tour dates yet. Not one. The robot did its diligence, came back, and basically said "I built you a lovely map of vibes but the actual bands part is aspirational." Respect for the honesty, mild personal humiliation for me.
The quotations around things that don't require quotations are starting to pile up. Some em dashes in there, but that's actually on brand.
**Round two: same request, but October instead**, because October is closer and therefore bands might have gotten around to telling anyone where they're playing. This is where things got good. Turns out Jason Isbell is doing a three-night stand at the Ryman. Turns out Old Crow Medicine Show is rolling through small-town Pennsylvania. Turns out Jeff Tweedy is doing a solo lap through Ohio with his kid's band opening. Carbon Leaf — a Virginia band, mind you — is out in Kansas City and Chicago being Virginian at people. Real dates. Real, bookable, small-venue shows. I felt like a guy who'd been handed an actual treasure map instead of a napkin drawing.
"Real dates. Real, bookable, small-venue shows." is an AI tell - one of the first in the text thus far.
**Round three: forget the loop, start on the West Coast and work back home.** This is where the trip stopped being "a nice drive" and started being "a bit." It turned out Iron & Wine — not on my original list, but close enough in spirit that I'm retroactively adopting him — is doing a whole honest-to-god coast-to-coast October tour, Seattle to South Carolina, and I basically hitched my itinerary to his tour bus like a barnacle. Which was efficient! It was also, in retrospect, a little much — I looked at the finished draft and realized I'd built a road trip that was 70% one guy's calendar. Bold move for a project explicitly premised on my loving lots of different artists.
The analysis part of this segment was pretty dumb, if we're being honest. And again with the extraneous quotations.
**Round four, the correction: no more than two shows per artist, please.** This is the "actually think about what you're doing" round, and it's the one that stuck. Rebuilt the whole thing with real variety — Carbon Leaf, Andrew Duhon, Iron & Wine, Shovels & Rope — capped at two apiece, no single act allowed to hog the whole trip like it's their solo project. There were still a couple of honest gaps (sorry, Asheville) where nobody's calendar cooperated.
Fine. No complaints.
**Round five: "I'll fly for anything over 400 miles," I said,** like some kind of aristocrat, and this single sentence fixed everything. Suddenly the Isbell/Ryman residency — previously geographically impossible without a week of pure interstate suffering — fit right into the middle of the trip. Memphis, which had been sitting there as a sad empty stop with no band attached, got adopted by an Iron & Wine date that the flights made reachable. Turns out the solution to "my itinerary doesn't make sense" was, the whole time, "money," which, yeah, tracks.
So here's where we landed. Five artists, capped at two shows each, three flights for the genuinely brutal legs, and a route that goes from a Virginia band playing a Seattle dive bar clear across the country back to actual Virginia. I did not confirm any of this personally, so don't yell at me if Carbon Leaf cancels — yell at the robot, or better yet, yell at Ticketmaster, the true villain in every version of this story.
Closing stronger that we opened, for sure.
---
Five artists, no repeats beyond two apiece, one month, roughly forty thousand emotional dollars in gas and airfare. Wager accordingly.
The last bit doesn't make a ton of sense. Forty thousand should be a number, not text. I don't know what emotional dollars are. Meh.
To sum up, I think the final product - the itinerary - is quite good, and something that would've taken me or any other human a long, frustrating time. That's a good use of the technology. The writing is okay, and if you weren't picky about your words, you might be okay with it. That's lame if you're looking for a shortcut for a term paper or a training manual, and it's dangerous - at least professionally - if you're putting together a legal brief or an op-ed. Or targeting attacks on Iranian schools.
I happen to be picky about my words, as I suspect most of us are, so this isn't good enough for me. It lacks a bit of heart, even as Claude tried to emulate our tone. And there are bits of off-sounding language that sours the whole thing for me as a reader. In the end, I think AI's ability to fake human writing is a technological marvel, but it's not nearly the most interesting nor I think enduring capability AI has. That comes from its ability to synthesize and make sense of massive amounts of information, draw inferences, and generate logical data quickly, and reasonably accurately.
Frankly, that's good enough to be really compelling. Shame the foofs that have a stake in AI's success can't just focus on that instead of moonshot claims about rendering human work redundant.
Posts
