What they did
Reece Rogers, a WIRED software reporter, spent a day running six always-on agents on his iPhone through everyday jobs: Meta's Muse, OpenAI's Dots, Instinct, Claude Duet, Gemini Spark and Grok Bot. One job was lunch. He asked each agent to build a takeout order from a place near the office that he could walk to and pick up, covering something meaty plus strong coffee for himself and a vegan Vietnamese meal for his boss. He committed to ordering from whichever agent did best. The agents browsed restaurant sites and delivery apps such as DoorDash and Uber Eats to put carts together.
How it went
Every agent landed on one of two nearby Vietnamese places. Dots scored 5 of 5: it picked the cheaper, tastier one through the restaurant's own site, swapped soy sauce for fish sauce on the boss's rolls, and came to $30.57 before tax. Rogers placed that order. Muse and Grok Bot built similar carts at about $43 through DoorDash. Claude Duet asked whether to use DoorDash or Uber Eats and came to $49.25, and Instinct was pricier still. Gemini Spark failed by giving the boss a single spring roll.
Worth knowing
This was one test in a single-day comparison, not long-term use. Delivery apps hide tax and fees until checkout, so cart totals the agents report are not the final price. Dots costs $100 a month, the most of the six.