ChatGPT agent books a Houston business lunch
Evan Sunwall reports a successful Brennan’s lunch reservation after slow research, a changed time, allergy entry and human contact-detail takeover.
nngroup.com
Original prompt
Shared by the source author. See source notes for context.
Find a restaurant suitable for a business lunch near 3100 Travis St, Houston, TX 77006 for next friday at noon.
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source author
- Evan Sunwall
- Original account
- Read the source ↗
- Published
- Source link checked
- Reported result
- Sunwall reports a confirmed 12:15 reservation after noon was unavailable, with successful allergy entry and substantial delays.
- Human involvement
- Sunwall chose the restaurant, supplied guest and allergy details, entered personal information and inspected the result.
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
- Request a business lunch search and review the restaurant candidates.
- Choose Brennan’s, supply guest count and add the dietary restriction.
- Let the agent select an available time and enter the allergy note.
Evan Sunwall tested ChatGPT agent with a deliberately brief request for a business lunch near a Houston address. The agent searched for restaurants and returned three candidates. Sunwall chose Brennan's, supplied a party of two and added a shellfish allergy.
The agent navigated the booking interface, struggled with the time selector and initially clicked the wrong time. It corrected that mistake, then found noon unavailable and selected 12:15 while warning about the change. It entered the allergy through the dietary controls and added a written note.
Sunwall supplied personal details through a takeover step. The account reports that the reservation was ultimately submitted and confirmed. The process took enough time that the website's hold window technically expired, although the booking still went through.
The successful outcome came with weak clarification and a high interaction cost. The agent began searching without asking what made a venue suitable or what the budget was. It spent six minutes researching and eleven minutes navigating booking forms, much longer than Sunwall's own comparison. This was one completed reservation with human choices and intervention, not evidence that the agent consistently saves time or that a shifted reservation time is always acceptable.
Prompt
Original prompt
Source: Blog · Evan Sunwall
Frequently asked questions
Did it reserve the requested noon time?
No. Noon was unavailable, so it selected 12:15 and warned the user about the change.
Who entered personal information?
The source describes human takeover for the remaining personal contact details.
Was the workflow faster than manual booking?
No. The author reports eleven minutes navigating forms versus roughly two minutes manually for the same flow.
More details
Remaining steps
- Take over for personal details and inspect the confirmed reservation.
About this account
The agent searches restaurants, operates booking controls, corrects the time selection and enters the allergy note.
Firsthand primary account opened and read; original publication date checked in primary metadata or visible timestamp. Outcomes are author-reported, not independently reproduced. One author-observed reservation; no general speed or allergy-safety guarantee. Original prompt excerpt reproduced exactly from the source; excerpt only, not the complete prompt.
Credit: Evan Sunwall
Source published
Added to Agents In Use
Source checked
Tools and topics
ChatGPT agent browser · Restaurant booking website
Travel · Reported deployment
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.