← Field Notes
SEP 25 · Clipped · via Simon Willison ApprovalConfidence SignalingPermissions

Meta's new agent looks safe because it asks before it acts

The confirmation prompt makes the agent feel safe. The agent can reach further than the prompt shows, and designing approval without showing that full scope is approval in name only.

Machine summary of the source

Meta has shipped what John Gruber calls the first major consumer general agent, built into its Ray-Ban smart glasses. Simon Willison flags a sharp editorial point from Gruber: the product's UX frames the agent as low-stakes because it asks permission before doing things. That framing may comfort users without reflecting the actual reach of what the agent can do. This gap matters for designers. A confirmation step makes an agent feel contained. The confirmation only covers what the human can see in the moment. If the agent has broad access behind that one prompt, the step does less than it appears. The human approves the surface action and may not realize the deeper access that comes with it. The design tension is whether the human reviewing a confirmation has enough information to judge what they are agreeing to. Gruber's framing is a warning: perception of safety and actual safety can diverge, and UX is the lever that moves both.

The summary above is generated; the note at the top is the editorial judgment. Primary source ↗