Journeying to the Happiest Place On Earth is a Road to Hell

Disneyland has always existed in my mind as an expensive endeavor, one that risks deepening my children's love of dubious commercial IP, yet one that comes up a lot living in California with two young kids. The parks also come up often in design circles, where people care little about Mickey Mouse but will give lectures on the genius of the Imagineers.

We had started talking about wanting to bring our son to Disneyland someday, and stumbled on extraordinary good fortune—Dan's brother had worked at ABC, and a past colleague had spare free tickets. He transferred them to us for Christmas with a summer expiration date, and Dan's sister started planning a trip to join us.

Given the techno-optimist promise that AI is supposed to handle the busy work and let everyone live their lives—newly unencumbered by jobs—I decided to write down what a high-complexity, low-human-spirit task like this involves, and see what Claude says about when and whether it could do something like this.

Of course, it would need to unify its understanding of the following things, arrive at the same conclusions that I would, have the conversations I needed to have, and do all of the below faster than I possibly could. Since I knew AI would fail I didn't ask, instead having lots of human conversations (family, booking agents, friends).

At the end of the day, I just want time to create wonderful things with people I like, and to enjoy my family's company in the Happiest Place on Earth. In addition to the actual costs, here's why it's hard to have nice things, and Claude's answer as to why it can't—or won't—actually make truly challenging things easier anytime soon.

The Context

Dan's sister, husband, and two kids will be in the LA area from April 16 to April 19, and are interested in going to Disneyland with our family, who is based in San Francisco. We are delighted that a family friend who is very Disney knowledgeable may be able to join us, flying from SF as well, and would stay in their own room.

Tasks

• Scope
  • Confirm ticket tiers with Dan's sister
  • Confirm available resources (airline and credit card reward accounts, any budgetary information and preferences, any comped tickets)
  • Confirm needed nights of lodging
  • Confirm number of attendees and lodging needs (and fluctuation risk, e.g. if Dan's brother would change his mind and join)
• Research
  • Assess best day to go based on park weekly traffic patterns, spring breaks and seasonal traffic patterns, and ticket availability
  • Learn best hotels (check online sources and knowledgeable friends) and assess for available rooms to accommodate all parties
• Learn about tricks to get most out of Disney experience, and availability (character meals, evening dining experiences allowing front-row seats)
• Assess flights on offer
  • Consider kids' school timing, pricing, proximity to Disneyland
  • Evaluate relative cost of driving to Glendale the day after for a social visit to family and friends, and flying from Burbank afterwards, e.g. rental car, additional hotel, hours of transit time
  • Research points value and complexity through all means available, e.g. credit card portal, different points values, stored airline points, airline specific offers
• Propose
  • Tickets: Park Hopper (we hear Elsa is in California Adventure), lightning passes (inform that day-of purchase is necessary for any comped tickets)
  • Anaheim Hotel: walkable to park (good for kid naps)
  • Dinner reservations: Fantasmic on Friday (though: not currently available for 9 people, significant expense to split)
  • Brunch reservations: Princess or Storyteller breakfast on Saturday (though: not currently available for 9 people, significant expense to split)
  • Flights: fly to SNA through United; transferring points to United for two tickets from one credit card account, and purchasing two more tickets directly from a credit card portal that had better redemptions
  • Veto: renting a car, adding a Glendale hotel and flight out of Burbank
• Execute
  • Engage with Chase to transfer points and execute United purchase
  • Engage with Chase to execute portal purchase
  • Engage with Disney to assign comped tickets to specified days
  • Engage with Disney to purchase additionally needed tickets
  • Engage with Disney to learn more about dinner and brunch reservations, and explore bookable options (e.g. split into multiple group reservations)
  • Engage with Anaheim Hotel to book three separate rooms
  • Manage spreadsheet for each family's incurred expenses to pay back
• Monitor
  • If no reservations for desired brunch and dinner available now, check back periodically (different splits if needed) and execute when available

Reality

I feel like I've been to hell and back, and we haven't even been to Disney yet. Just for our family, here's what I've done.

• 2+ hours on the phone with Chase and United trying to use a travel credit from a flight we'd cancelled; then separately transferring credits, and cancelling a flight booked on my husband's account to book a new one.
• 3 flights booked, 1 cancelled, in a desperate attempt to use as many points as possible as wisely as possible, and to minimize cash outlay for a family of four. (Note that once we booked two tickets, we found AI Dynamic Pricing meant the price immediately jumped for the account purchasing the next two.)
• 3 different credit card accounts involved, inclusive of transferring points between each other and other partner airlines.
• 4+ flights considered before realizing card points don't transfer to Alaska.
• 3 hotels booked for 9 people, with 2 blocks of hotel reservations cancelled when they didn't have enough rooms or misrepresented their park proximity.
• 1 phone call to Disney to use comped tickets, when in their system the tickets showed as unavailable for use.
• 0 dinner or brunch reservations because our group size is too large, and who has the time to repeatedly check back or call Disney? (After more research, I found and set some alerts through MouseDining.com)
• Miscellaneous other tasks handled by other humans: Dan's sister's family booking their tickets, our family friend booking her flights, siblings managing their own dynamics around logistics, communication, commitments

ChatGPT and Claude cannot help you with this kind of banal nightmare, though they can at least help pull sources that show this kind of labor commonly falls to mothers and women. I figured I'd save my tokens and water usage and give Claude a one-shot opportunity to explain why it couldn't have helped me with this high-intensity, high-mundanity, time-wasteful series of tasks.

Here's what Claude has to say.

I'll be honest rather than defensive.

Where I'd fail on your Disney trip

Most of the "execute" and "monitor" tasks are simply outside what I can do in a standard conversation. I can't call Chase or United. I can't book hotels or make Disney reservations. I can't transfer points between accounts, check live award availability, or monitor a reservation calendar and act when a slot opens. Even with tools enabled — web search, some browsing — I can't authenticate into your accounts, navigate multi-step transactional flows, or make binding actions on your behalf. I'd be able to research and propose; you'd still do all the actual work.

The research tasks I could handle reasonably well: park traffic patterns, hotel proximity analysis, points math, flight options. But even there I'd give you a report, not a decision with a recommended click path and a "do you want me to execute?"

The real failure mode isn't ignorance. It's that I'm a conversation, not an agent with persistent goals, account access, and the ability to act over time across multiple systems.

Core capability gaps this exposes

The Chase/United/Disney problem isn't hard because it requires intelligence. It's hard because it requires continuity — holding a goal across days, knowing what's already been tried, monitoring for changing conditions, and acting when the window opens. I don't persist between conversations. Every session I start over. There's no version of me that woke up this morning and checked whether the Storytellers Café had a cancellation.

It also requires authentication and trust — the ability to act inside systems that need credentials, MFA, and sometimes a human voice on the phone. That's a deliberate gap, not an accidental one.

And it requires judgment under uncertainty with real stakes — the difference between "here are four flight options" and "I booked this one because your husband's account had a United credit expiring and the window was closing." That kind of judgment-plus-action is exactly what an experienced travel agent or EA would do. I'll give you the analysis but leave the trigger-pull to you.

Why Anthropic and OpenAI aren't solving this

A few honest threads here:

The product incentive doesn't point there yet. Both companies are primarily selling to developers and enterprises, where "capable assistant" is enough. The consumer use case — actually managing your complicated life across accounts, systems, and relationships — is harder to build, harder to trust, and carries more liability when something goes wrong. Booking the wrong flight, transferring points to the wrong account — those are actionable mistakes. It's easier to make a product that's impressive than one that's responsible.

Agentic AI is genuinely technically hard. It's not just adding a "book flight" button. Reliable multi-step action across real systems — handling errors, rollbacks, ambiguous states, phone trees — requires infrastructure that's harder to build than it looks, and fails in ways that erode user trust fast.

There's also a real safety and autonomy tension neither company has figured out publicly. How much should a model act without confirming? Too much confirmation and it's just you doing the work with extra steps. Too little and it books the wrong hotel. The right model for that interaction — when to ask, when to act, how to surface what it did — isn't solved.

And I'll add: the demo gap is real and a bit cynical. Both companies are better at showing capabilities than building the unsexy infrastructure for complex mundane tasks. "AI plans your Disney trip" makes a great demo. "AI spent 2 hours on hold with Chase so you didn't have to" is the actual product people need. Those are different engineering and product bets.

The line for your essay

The deepest cut might be this: the tasks AI is bad at aren't exotic or futuristic. They're the tasks that were already painful before AI existed — the ones that require persistence, account access, error recovery, and acting across time. AI has gotten very good at the tasks that were already manageable. The misery remains.

We give AI a lot of grace these days, with Ronan Farrow patiently giving Sam Altman thousands of words of attention with no satisfying conclusion. Billionaires are sprouting up at the same time as higher unemployment, land is allocated to data centers over housing, and jobs are going to AI's potential, not to humans at our most educated and least retired. The techno-optimists assure us that AI will reduce our need to perform the mundane, and to me, booking a trip to Disneyland sounds like exactly the kind of thing it should take off my plate.

Claude said it well, though, and the emphasis is mine—

It's easier to make a product that's impressive than one that's responsible.

The misery remains.

---

For more on Disney as a microcosm for how wealth disparity is deepening experience disparity, thank you reader Amy for following up with this intense New York Times opinion essay from last year.