How to use Muse to reorder a family lunch
Muse repeated a DoorDash lunch after chat approval and texted pickup details; Claire Fang’s comparison exposed missing history tools and a missed promotion.
linkedin.com
Original prompt
Shared by the source author. See source notes for context.
Can you order the same thing on DoorDash as we ordered last Saturday?
Reach people exploring AI agents
See placements →What happened
Fang reports a completed Muse reorder and requested confirmation text; other tested routes showed slower recovery, missing history tools and a missed promotion.
Reported by Claire Fang — LinkedInHow the agent did it
- Fang asked each experience to repeat the family’s previous Saturday DoorDash order.
- Muse requested login takeover, recovered order details and presented them in chat for human approval.
- After ordering, Muse texted the requested pickup confirmation; the comparison exposed history-access and promotion failures elsewhere.
Where a person stepped in
Fang handled authentication, reviewed and approved the order, and explicitly requested the text to her daughter. Comparative findings come from this one family workflow.
Claire Fang tested four AI shopping experiences on her family’s usual Saturday Subway lunch. The request depended on retrieving a previous DoorDash order, which made account history more important than generating a new meal suggestion.
Muse handed the browser to Fang for login, recovered the relevant order and brought its details back into chat for review and approval. After the order was placed, Fang asked it to text the confirmation to her daughter for pickup. That completed handoff is the strongest reported result in the comparison.
The ChatGPT Work browser agent also reached the correct cart, but slowly recovered from a stalled order-details page by returning to history. Its review required another browser takeover. DoorDash’s direct ChatGPT integration could not retrieve past orders, despite supporting other grocery capabilities, and fell back to browsing.
DoorDash’s native assistant produced the sandwiches but missed a current thirty-percent promotion. Fang therefore distinguishes a smooth conversation from the correct economic result. This is one first-person comparative case, not four independent use cases or a general benchmark. Login, order approval and the daughter’s pickup instructions remained human-directed.
Prompt
Original prompt
Source: LinkedIn · Claire Fang
Frequently asked questions
Why did the direct DoorDash integration fall back to browsing?
It could not retrieve past orders for this request, despite supporting other shopping functionality.
Where did Fang review the Muse order?
Muse brought the relevant details into chat for review and approval.
Did smooth interaction guarantee the best price?
No. The native DoorDash assistant missed a current thirty-percent promotion in this test.
More details
About this account
Fang asked each experience to repeat the family’s previous Saturday DoorDash order. Muse requested login takeover, recovered order details and presented them in chat for human approval.
Full primary comparative article opened. Original shared prompt. One canonical source counts once. Meta Muse is primary because its completed reorder and pickup-message handoff are the central result; alternatives are scoped comparison evidence. Original prompt excerpt.
Credit: Claire Fang — LinkedIn
Source published 28 Sept 2026
Source checked 14 Oct 2026
Tools and topics
Meta Muse · DoorDash · ChatGPT Work browser agent · Text messaging
previous-order retrieval · chat approval · promotion comparison
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.