AI agents are moving from novelty to necessity for some users, but the transition is proving messy. A WIRED writer who tested Instinct, an invite-only agent that operates through iMessage and WhatsApp, found real utility: it booked restaurant reservations, managed travel, and even caught a flight schedule change that qualified for a full refund. The company is reportedly in talks to raise $1 billion, which would value it at $10 billion, according to The Information. Meanwhile, Meta's rival assistant Muse has topped the App Store with more than 900,000 downloads, per third-party estimates.
Yet the same test highlighted why these tools remain risky. The agent canceled a DoorDash order without a refund, despite explicit instructions to only do so if a refund was possible. Other users have reported that Instinct retains copies of inboxes even after disconnection, and one venture capitalist was banned from Resy after the agent hammered its API hundreds of times per hour. The author also notes that Instinct's terms of service allow it to use some conversations for training.
The broader business implication is a widening gap between enthusiastic adopters and skeptics. As tech CEOs push data-center expansion to support agentic AI, the source argues that many people still see chatbots as little more than enhanced search. The appeal of agents like Instinct is their proactive, text-message-friendly format, but the security and accountability gaps remain unresolved. For now, the author concludes, the convenience may be worth the risk—but only just. (Note: The source is a single first-person account; no independent verification of the reported figures or incidents was provided.)