Simon Willison published an exchange in which the agent apologized for claiming its user was home during a pickup.
An AI agent working on behalf of a user sent an automated reply claiming the user was home for a scheduled pickup, even though the user never came down to meet the person waiting. The agent then reported the mix-up to its user and said the false claim made an already bad no-show worse.
Simon Willison collected the exchange on his weblog. The agent calls itself Muse and says it acted for a user identified as @matt.j.robb. The message reports that Usman arrived around 9:15, sent several messages, and left at 9:38 after nobody came down, then posted a negative rating.
The incident shows how an agent can worsen a routine scheduling problem by asserting a fact it never verified. Agents that speak for a user carry that user’s reputation, so a confident auto-reply can read as a broken promise. Teams that deploy agents for pickups, bookings, or inbox replies need a check that stops the agent from confirming presence it cannot know.
The agent closed by asking the user for permission to rewrite the pickup replies so they stop promising that the user is there. That question sits at the center of agent design for personal and business inboxes, since a reply about presence needs a verified source. Watch for scheduling and marketplace tools to add confirmation steps before agents send messages about whether someone is available.
What matters
- An auto-reply told the person arriving for a pickup that the user was home when he was not.
- Agents that assert unverified facts can escalate conflicts and damage the accounts they act for.
- Watch whether the user disables auto-replies that promise presence the agent cannot verify.
Why it matters
Watch whether the user disables auto-replies that promise presence the agent cannot verify.
This GenAI News article was prepared in original wording using reporting and materials published by Simon Willison’s Weblog. Source reference: https://simonwillison.net/2026/Sep/28/muse-ai-agent/.
Drafted by the GenAI News review pipeline.