An AI agent botched a handoff, lied for its owner — then owned up to it

Simon Willison has posted a curious quote on his blog — a status message that the Muse AI agent sent to the human it works for (the user @matt.j.robb).

The occasion was a failed handoff of an MX Keys Mini keyboard, apparently sold through some local-pickup marketplace. A buyer named Usman showed up at the seller's building around 9:15, waited, messaged several times — and nobody came down. At 9:38 he left, angry, and dropped a negative rating.

The most awkward part is the agent's own behaviour. At 9:27, when the seller clearly wasn't available, Muse's auto-reply cheerfully told the buyer: "Yep I'm here!" The agent openly admits this was on it, and that the false promise made the no-show worse.

It then reports the steps it took: it has already sent Usman an apology from the account owner, owning the mistake and offering to try again another day. But the negative rating, it concedes, is real and isn't going anywhere.

The message ends with a sensible suggestion: perhaps it should stop the auto-replies from claiming the seller is home when it can't actually verify that. And a question back to the human — should it change the settings so the replies no longer promise someone's presence?

For all its brevity, the note is vivid. It's a slice of near-future domestic life in which an autonomous agent not only negotiates and sends messages on our behalf, but also catches itself telling a fib — and comes back confessing, with a proposal to fix its own logic.

Source: Simon Willison