WIRED's one-day agent test gives Instinct low marks and its Most Annoying label
WIRED · ONE-DAY AGENT TEST
MISSED CONFLICT
Instinct also gets WIRED's "Most Annoying" label
2/5
schedule conflicts
missed planted overlap
4/5
insomnia check
reasonable revision offer
1 · 2 · 2 · 3 · 4
five task scores
six agents, one reviewer
WIRED · 27 Oct 2026, 5:30 pm MYT · 24h, personal Gmail + Calendar, limited sample
WIRED's Reece Rogers spent 24 hours testing six always-on agents across five tasks. Instinct scored 1/5 for the morning briefing, 2/5 for lunch, 2/5 for schedule conflicts, 3/5 for vacation planning and 4/5 for the insomnia check, then received WIRED's Most Annoying label.
What happenedRogers ran Meta's Muse, OpenAI's Dots, Instinct, Anthropic's Claude Duet, Google's Gemini Spark and xAI's Grok Bot through one personal day of tasks. The test used his personal Gmail and Google Calendar; OpenClaw, Poke and Hermes were not tested.
Instinct's scoresThe morning briefing scored 1/5 as a bland, hard-to-measure experiment. Lunch scored 2/5 after a Freshroll cart reached $53.73 before tax and fees in the review. Schedule conflicts scored 2/5 because Instinct missed the planted email-calendar overlap, while offering to run future checks. Vacation planning scored 3/5 as decent and helpful, and the insomnia check scored 4/5 for a reasonable offer to revise the China itinerary by morning.
Why it mattersThe sharpest failure is proactive reliability: Instinct did not connect a self-sent email reminder with a conflicting calendar event. Rogers also called the texting-only experience a deal-breaker, saying he felt too removed from the agent's activity and was annoyed by emoji reactions after the missed conflict. He notes that other WIRED writers have praised Instinct.
Evidence limitsThis is one person's five tasks over 24 hours, not a controlled repeated benchmark or a test of every major capability. The six-product field is not exhaustive, and the prices/access described are the reporter's context, not independently verified current plans. The $53.73 lunch subtotal is a historical cart from this test, not a current restaurant price.
Practical checkTreat this as a prompt to verify your own conflict detection and inspect consequential actions before relying on any agent unattended. That caution applies across the tested agents, not only to Instinct.
Sources WIRED · Reece Rogers