Two personal AI assistants, OpenAI's Dots and Meta's Muse, were tested in a practical scenario: planning a hypothetical trip to the WNBA Finals in Atlanta to see the Golden State Valkyries, organizing a purple-themed brunch watch party, and producing souvenir programs. Both agents handled multi-step tasks well, but showed shortcomings in factual consistency, final editing, and executing real purchases.
Why this matters
AI companies are competing to build digital assistants that users rely on for shopping, travel planning and other everyday tasks. A key question remains whether these agents can match or exceed human performance ('the human benchmark') and whether users will trust them with meaningful account access and financial transactions.
The test assignments
The author asked each assistant to:
- Search for flights and tickets for Game 3 in Atlanta.
- Organize a purple-themed brunch watch party with recipes and invitations.
- Create a 1980s-style WNBA Finals souvenir program and turn photos from Game 1 into a keepsake.
- Help shop for Valkyries merchandise, which evolved into buying WNBA trading cards.
Strengths observed
Both agents surprisingly handled complex, multi-step tasks.
- Scheduling and ticketing: both noted that Game 3 had been moved up by a day and that tickets were not yet on general sale; available seats were on the resale market and likely inflated.
- Flights: both found good flight options.
- Ticket presale path: Muse opened a browser and allowed the author to enter information to attempt joining a ticket presale. Dots flagged a scheduling conflict derived from the author’s calendar related to an enrollment fair.
Creative output—and its limits
Each assistant could transform broad instructions into elaborate deliverables, but neither replaced a human editor.
- Brunch materials: Muse produced a colorful 10-page recipe booklet (including ube waffles and lavender-honey butter); Dots produced a more utilitarian 8-page packet with detailed cooking instructions and attention to dietary preferences. Muse’s shopping list requested 12 eggs while its recipes required 13—an example of small but consequential inconsistencies.
- Invitations and design elements from both agents required aesthetic and language edits before being shareable.
- The generated menus were far more elaborate than what the author would realistically prepare for a watch party.
Souvenir programs and photo keepsakes
Both agents created retro-style programs:
- Dots produced a 38-page vintage program that was visually engaging, with player photos, graphics and extensive statistics.
- Muse produced a 35-page version that was less attractive but included playful touches like a fictional computer ad and an ad for the imagined brunch.
Both programs contained questionable statistics; Muse later sent an unsolicited update correcting several factual errors. For the photo keepsake, Dots produced a 12-page version in 47 seconds, while Muse took considerably longer but suggested an interesting spread spotlighting different generations of Valkyries fans.
Where execution fell short
When it came to actually completing purchases, both agents hit limits.
- The author chose Target for purchasing trading cards (using a store credit card that provided a 5% discount and free shipping).
- Muse attempted to complete the purchase but stalled at Target’s sign-in page.
- Dots directed the author to the Target app, where account and payment details were already saved; despite being less automated, that approach proved faster.
These outcomes show that even when agents support purchases with user approval, technical and permissions-related hurdles often impede final completion.
The trust problem
Both agents could have done more if the author had granted broader access to accounts and payment methods. The author expressed discomfort with authorizing large-dollar automated purchases or entrusting consequential decisions to the agents at this stage.
Bottom line
Dots and Muse are useful for research, planning and handling legwork, but they are not yet replacements for human judgment, final editing or sensitive financial operations. The author did not travel to Atlanta or host the brunch due to other commitments, but did end up buying three boxes of WNBA basketball cards—ostensibly for research.
