I Tested Every Major AI Agent for a Day. Here’s the Best One

A hands-on WIRED test found OpenAI Dots performed best overall among six always-on AI agents, with Meta Muse a close second and free; Dots costs $100 a month, while tested subscriptions ranged from $20 to $30 monthly for several other agents. Results were uneven: only two agents caught a calendar/email conflict, and Gemini Spark failed the lunch-order task. The review notes improving reliability but flags privacy and security risks from connecting agents to personal data, and recommends most users wait for cheaper access to Dots.

Analysis

The investable issue is control of the shopping funnel, not which assistant won a small product test. If consumers delegate product discovery and checkout, retailers risk losing search, recommendation, and retail-media economics; an agent that routes a shopper across stores can make price and availability more comparable. Walmart’s openness to third-party agents could attract incremental transactions, but only if it retains checkout, customer data, and enough attribution to monetize them. Amazon’s tighter control may protect its funnel in the near term while risking reduced reach if shoppers increasingly use external agents. Neither outcome is established by this anecdotal review.

META has a distribution option: a free, widely accessible assistant could seed commerce use and later support monetization, but privacy concerns and unreliable execution could constrain sensitive permissions. GOOG’s weak showing in this narrow test is not evidence of a broad product or earnings disadvantage. Likewise, subscription pricing and polished interfaces do not establish durable willingness to pay; agent usage could add inference costs before generating attractive revenue.

Near term, this is not an earnings signal. Over 1–3 months, watch retailer policies, agent referral/checkout conversion, and evidence of repeat use. Over 6–18 months, the structural question is whether retailers can capture agent-originated demand without surrendering customer ownership. Privacy incidents, unauthorized actions, or poor task completion could slow adoption and reverse the narrative. The review is too small and too early to justify a directional trade on its own.

Key Decisions for Investors

• No immediate position from this article alone. Treat product rankings as weak evidence; do not extrapolate them to META, GOOG, AMZN, or WMT revenue forecasts.
• Put WMT versus AMZN on a conditional relative-value watchlist, not an active pair: consider long WMT / short AMZN only after evidence shows agent referrals produce incremental Walmart conversion while Amazon’s closed approach loses measurable traffic or share. Falsify the setup if agent-originated orders remain immaterial or Walmart cannot retain checkout and attribution.
• For META, monitor active agent users, repeat task completion, commerce conversion, and privacy-related permission changes. Adoption without measurable monetization—or a material trust incident—would weaken the option value; do not treat free downloads alone as a catalyst.
• Track agent error rates, unauthorized actions, and customer-support or refund friction alongside retailer integration announcements. A visible deterioration in trust would be a near-term adoption risk; successful transactions with repeat usage would be stronger validation than launch claims.