A Shopping Agent Revises Room Designs to Fit Budgets and Delivery Needs
OpenAI’s Vesper & Alder demo presents shopping as a series of connected decisions: a shopper can revise a reading-nook selection to meet a $250 budget without abandoning the design goal, then consider a throw as a separate add-on. The flow also lets the shopper correct the fulfillment choice, switching from store pickup to home delivery during the order process.
A budget constraint changes the selection, not the project
The shopping interaction is organized around a specific goal: make a bare room corner warmer as a reading nook. After Alex asks for ideas and uploads a photo, Alder proposes a cream chair, a light oak table, and a warm linen-shade lamp. The app pairs those product choices with a generated preview showing them in the room.
That pairing gives the shopper something concrete to respond to. Alex does not have to start over when the first set comes in over budget: the initial three pieces total $287 before tax, and Alex asks to keep them under $250. Alder swaps in a Lummi cream chair and updates the visualization. The app presents the original design at $287 beside the revised one at $237, both before tax.
The revised set preserves the reading-nook idea while changing a product to meet the price constraint. Alex can compare the two room previews and decide whether the less expensive selection still works in context. In the exchange, the image is not just a mood board: it appears alongside the product list and changes when the selection changes.
The next recommendation is optional—and separate from the budgeted set
Once Alex is ready to check out, Alder offers a Natural Stripe Throw as an additional way to add warmth. The screen prices it at $17.50, reduced from $35, and offers a room preview. Alex agrees to add it.
The offer is distinct from the three-piece set Alex had just brought under budget. That distinction matters to the flow: the throw is presented as a further choice at checkout, rather than silently folded into the revised selection. Alder asks, and Alex decides.
The displayed amounts do not line up cleanly. The offer screen shows $277.41 including tax, plus $17.50 for the throw; the subsequent order-review screen shows $277.41 for “Your nook, with the throw.” The figures appear together in the checkout flow without an explanation of how the added item affects the total.
The purchase can still change after product selection
The order review shows pickup at Berkeley selected, with the address at 1100 Fourth Street and a note from Alder that pickup takes five days. Alex realizes the order was meant for home delivery instead and asks for help. Alder requests the street address, city, state, and ZIP code, then displays the supplied San Francisco address for review. Alex confirms it, and Alder says the order has been updated for delivery on October 9.
That correction is a different kind of change from swapping the chair. The earlier revision altered the contents of the proposed nook; this one changes how the order will reach Alex. In both cases, the interaction continues from the shopper’s stated preference: first a price ceiling, then a delivery address. The address is shown back to Alex before the update, and the throw is offered for approval before it is added.
Taken together, the screens and exchanges show a shopping flow in which product selection, a visual preview, an optional add-on, and fulfillment are handled as connected decisions. The shopper can revise the set without abandoning the design goal, consider an additional item separately, and correct the delivery choice during the order process.