One Writer Put Six AI Agents Through Their Paces for a Day - Here's How They Stacked Up

AI agents have come a long way from the days when they simply didn't work. A WIRED software reporter who once complained endlessly about their uselessness spent 24 hours testing six always-on agents - Meta's Muse, OpenAI's Dots, Instinct, Anthropic's Claude Duet, Google's Gemini Spark and xAI's Grok Bot - to see which ones actually deliver.

The agents vary widely in price and access. Muse is free and has topped smartphone download charts since its September release. Instinct is also free but requires an invite code and runs through texts rather than an app. Dots sits behind a $100-a-month paywall inside ChatGPT. Gemini Spark is a $20-a-month beta within the Gemini app. Grok Bot costs $30 a month in its own app, while Claude Duet, the newest of the bunch, costs $20 a month as a wait-listed research preview.

Five tests measured proactivity, instruction-following and browsing: compiling a tech news briefing, ordering lunch for the reporter and a vegan boss, catching a calendar conflict against an email reminder, drafting a China vacation itinerary, and handling a middle-of-the-night request. Only Muse and Dots kept researching for most of the allotted half hour on the briefing task. Just two agents spotted the schedule clash. All six proposed identical China routes - Beijing, Xi'an, Chengdu and Shanghai - raising the specter of a new Lonely Planet effect.

Dots took best overall despite the steep price, with the writer suggesting most people wait for a cheaper tier. Muse finished a close second and was the easiest to set up. Claude Duet earned best interface, while Instinct took most annoying for its texting-only design and emoji reactions. The reporter cautioned that the software remains nascent and carries privacy and security concerns, noting one X user who connected agents to his bank account and Slack only to have his balance posted for colleagues to see.