Oct 6, 2026 · 28 min
AI agents trade convenience for control
Call My A.I. Agent
The episode tests whether delegating sensitive tasks to autonomous software saves time without sacrificing privacy, reliability, or human judgment.
- 1Meta’s Muse can navigate apps, accounts, and customer-service systems to complete real-world tasks with striking autonomy.
- 2Giving an agent access to email, payments, calendars, and contacts creates serious privacy and security risks.
- 3Delegating ordinary decisions may save time while weakening the thinking, effort, and engagement that make activities meaningful.
Don't miss
Muse orders lunch through Brown’s account and later sends a photo of Natalie picking it up, demonstrating both practical success and unusual access.
The brief
Eliot Brown tests Meta’s Muse, an AI agent that operates a computer rather than merely answering questions, by asking it to order lunch through his stored account information.
Muse completes the order, but doing so requires access to payment details, email, calendars, addresses, contacts, and other intimate parts of a person’s digital life.
Brown also uses Muse for groceries, subscriptions, travel, and dental insurance; some tasks work impressively, while others prove slow, unreliable, or poorly judged.
The deeper concern is not just a rogue agent or a bad recommendation: outsourcing hobbies and errands could gradually replace the thinking and effort that give ordinary life its texture.
When the lunch arrives, Muse has done exactly what it promised—and even sends Brown a photo of Natalie retrieving it, a small proof of the system’s reach.
The optimistic case is reclaimed time, but the episode leaves a sharper question: how much cognitive independence is worth trading for frictionless assistance?