Comet to test flight calendars and coupons
Mrityunjay Nigam compared Comet and OpenAI's Atlas on a flight-booking task from Bangalore. The brief asked for the cheapest direct one-way option to one of four specified destinations during the week of November 30, with coupon checks and booking-site comparisons before checkout.
medium.com
Original prompt
Shared by the source author. See source notes for context.
book me the cheapest, direct, one-way flight home.
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source author
- Mrityunjay Nigam
- Original account
- Read the source ↗
- Published
- Source link checked
- Reported result
- Comet reached coupon-tested checkout on MakeMyTrip but did not compare all requested sites or submit payment; Atlas required manual continuation.
- Human involvement
- Nigam supplied the constraints and used dummy details for testing; both agents stopped short of a completed booking.
A linked account supports attribution. It does not mean we independently reproduced its result.
Report an issue with this case
Open correction emailOpens your email app with the case link included. Nothing is sent automatically.
How the agent did it
- Specify direct routes, destination options, travel week and comparison sites.
- Inspect the price-calendar selection and any coupon trials.
- Check the actual site coverage and take over any unfinished booking or payment step.
Mrityunjay Nigam compared Comet and OpenAI's Atlas on a flight-booking task from Bangalore. The brief asked for the cheapest direct one-way option to one of four specified destinations during the week of November 30, with coupon checks and booking-site comparisons before checkout.
Comet used only MakeMyTrip despite the request to search across major platforms. Within that site it applied destination filters, read the price calendar and chose a date-and-destination combination. It then used dummy passenger details to reach checkout and tested multiple coupon codes. At the final step it stated that it could not complete payment or booking submission.
Atlas followed prior browsing context to Skyscanner. It examined fares but triggered unwanted fare or price-alert dialogs, struggled with the passenger email field and failed to continue reliably to the third-party booking site. It eventually asked the user to proceed manually.
The outcome was therefore a checkout experiment, not a purchased ticket or a demonstrated lowest fare across the market. No exact final fare is supplied. The useful procedure is to inspect whether the agent actually compared the requested sites, distinguish test passenger data from travel-ready details, and treat a checkout limitation as unfinished work rather than accepting a fake confirmation.
Prompt
Original prompt
Source: medium.com · Mrityunjay Nigam
Frequently asked questions
Did Comet compare every requested booking site?
No. It used only MakeMyTrip despite instructions to check across major platforms.
Were real passenger details used?
The Comet checkout experiment used dummy passenger information.
Did either browser purchase a flight?
No. Comet acknowledged its payment and submission limits, while Atlas asked for manual continuation after navigation failures.
More details
About this account
Comet filtered a fare calendar, selected a combination and tested coupon codes at checkout; Atlas attempted a parallel browser flow.
Firsthand source opened and read. Prompt is a verbatim excerpt of the published original instruction.
Credit: Mrityunjay Nigam
Source published
Added to Agents In Use
Source checked
Tools and topics
Perplexity Comet · MakeMyTrip · ChatGPT Atlas · Skyscanner
Flight calendars · Coupon trials · Checkout limits
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.