Real Agent Cases

Book a trip, flight, hotel and shuttle with ChatGPT Dots, and keep confirming the same booking (partial)

Posted
by neom (Hacker News)
Last checked
Method:
How we verify

Evidence: Described by posterThe person describes what happened in words, without screenshots or video.

PartialThe source says it only partly worked, or had a significant problem a person had to deal with.

What went wrong: The trip got booked, but the agent kept re-asking for confirmation and couldn't handle sign-in codes.

Outcome

neom spent two hours trying ChatGPT's Dots agent and grades it B-. Asked to figure out a trip, it did; neom booked the flight and hotel, and Dots noticed the shuttle hadn't been booked and offered to book it with the details they'd discussed. But after neom said yes, it went back to the site, re-checked the details, asked for confirmation, filled in the booking form and then asked again whether to book. It can sign in with passwords, but wouldn't fetch two-factor codes from email even with email access, and its permissions checker rejected broad rules neom wrote, such as proceeding with routine, reversible actions without asking. neom finds it faster to use regular chat.

Agent

ChatGPT Dots

Prompt Recreated prompt

The source didn't share its exact prompt, so we wrote this from what it describes.

Before you start, ask me 1–3 short questions about anything important that's missing below. Keep them quick. Then do the task, checking in with me at the key points.

Goal: Plan a trip with me and book the pieces I approve, including the transfers people forget.

What I can give you (ask for whatever's missing):
- Where, when and who's travelling: [DETAILS]
- Budget and preferences for flights and hotel: [DETAILS]
- Which accounts you can use to book: [DETAILS]

Rules:
- Propose a plan first: flights, hotel and how I get between the airport and the hotel.
- Once I approve a booking, ask for my OK once with the full summary, then book it without asking again.
- Check for anything I've missed, like transfers or seat or bag fees.
- If a sign-in needs a code, tell me straight away so I can enter it.
- If a site asks me to sign in, pause and let me do it myself (including any CAPTCHA or two-step code); never guess or store passwords.
- Ask for my OK before you pay, book, send, submit or cancel anything.
- Check facts that can change (prices, opening hours, dates, availability, rules) against the official source, and tell me anything you couldn't verify.

Output: A list of everything booked with confirmation numbers, plus anything still to do.

You'll know it worked when: the whole trip is booked, door to door, with one approval per booking.

Try this in:

This prompt has blanks to fill in, like [DETAILS]. We'll copy it instead of filling it in, so you can edit it before you send.

Opens in a new tab. We don't see what you send.

These links are only a shortcut. We're not affiliated with OpenAI, Google or xAI, and your results may not match the source.

Optional stronger version

Our rewrite with extra checks. The poster didn't test it.

Before you start, ask me 1–3 short questions about anything important that's missing below. Keep them quick. Then do the task, checking in with me at the key points.

Goal: Plan a trip with me and book the pieces I approve, including the transfers people forget.

What I can give you (ask for whatever's missing):
- Where, when and who's travelling: [DETAILS]
- Budget and preferences for flights and hotel: [DETAILS]
- Which accounts you can use to book: [DETAILS]

Rules:
- Propose a plan first: flights, hotel and how I get between the airport and the hotel.
- Once I approve a booking, ask for my OK once with the full summary, then book it without asking again.
- Check for anything I've missed, like transfers or seat or bag fees.
- If a sign-in needs a code, tell me straight away so I can enter it.
- If a site asks me to sign in, pause and let me do it myself (including any CAPTCHA or two-step code); never guess or store passwords.
- Ask for my OK before you pay, book, send, submit or cancel anything.
- Check facts that can change (prices, opening hours, dates, availability, rules) against the official source, and tell me anything you couldn't verify.
- Put every booking, time and confirmation number into one itinerary at the end.

Output: A list of everything booked with confirmation numbers, plus anything still to do.

You'll know it worked when: the whole trip is booked, door to door, with one approval per booking.

Before you finish: re-read my goal and constraints, check your answer against each one, and list anything you assumed or couldn't confirm.

Source

https://news.ycombinator.com/item?id=49904198

  • Book every restaurant for a Japan trip with MuseRecreated promptEvidence: Described by poster

    Muse booked every restaurant for the poster's Japan trip and corresponded in Japanese with a Hakone hotel to secure a dinner slot not offered on its website.

    Muse · Travel · Source date:

  • Hunt last-minute holiday flights with multiple agents (failure case)Improved promptEvidence: Described by posterFailure

    Claude Opus 4.5, Gemini 3 Pro and ChatGPT agent mode only matched a manual Google Flights search and were too slow as cheap fares vanished; the humans beat them by about 15%.

    ChatGPT agent, Claude Opus 4.5 (Antigravity), Gemini 3 Pro · Travel · Source date:

  • Search for cheap flights with ChatGPT agent (failure case)Improved promptEvidence: Described by posterFailure

    ChatGPT agent suggested flights costing $1,400 when the poster found an $850 option on their own.

    ChatGPT agent · Travel · Source date: