nytimes.com web signal

Meta Muse and OpenAI Dots Tested on 5 Daily Tasks by NYT

TL;DR

  • NYT's Brian X. Chen ran Meta's Muse and OpenAI's Dots through five everyday chores, from job applications to Michelin-tire price haggling.
  • Muse offers a free tier plus $20 and $100 monthly plans; Dots is included with ChatGPT Pro and Business Premium and runs on GPT-6 Astra.
  • Muse answered eight nanny postings and drew two parent replies in an hour, while Dots stalled on browser glitches during the same test.

Brian X. Chen spent his latest column at The New York Times pointing Meta's Muse and OpenAI's Dots at five everyday chores: restaurant bookings, workout nudges, job applications, daily meal plans, and price haggling. The agents handled complicated, multistep work well enough, he writes, that users "may never go back."

The job search produced the clearest split. Chen set the agents loose on nanny postings. Muse "successfully responded to eight postings in one instance, generating two parent inquiries within an hour," while Dots kept hitting glitches in its browser. On a reservation for Sun Moon Studio, a 12-seat Michelin-starred restaurant in Oakland, both agents failed repeatedly before Dots eventually reached a confirmation screen for a late-November table. On tires, both agents surfaced a competing quote and the review notes "finding $80 savings on Michelin tires," then requested a price match from a nearby shop.

The pricing and architecture diverge. Muse is a "free tier with usage limits, plus plans at $20 and $100 a month," and is the first agent covered by Stripe's warranties and return policy. Dots is a paid tier inside ChatGPT, powered by GPT-6 Astra, and "connects to more than 4,000 apps through plugins." Chen also set up a separate Gmail account for shopping, since the companies may review agent activity to improve the products.

Two of the AI practitioners on our radar passed the column around the day it ran.

Shared on Bluesky by 2 AI experts

  • Alex Hern @hern.bsky.social amplified

    @dlknowles.bsky.social

    I am just begging people writing about AI to think *one step ahead* and try to anticipate the actions of other humans. Three out of five of these AI agent "killer tasks" only work because *your* AI agent isn't yet up ag…

    View on Bluesky →
  • Joe Bak-Coleman @jbakcoleman.bsky.social amplified

    @matthewus.bsky.social

    Jfc the AI boosters who pass for journalists at the Gray Lady are getting desperate. Here, Brian Chen talks about 5 "killer tasks" for AI, all of which show more that he needs to examine his priorities rather than that h…

    View on Bluesky →