AIToday
Large Language ModelsSimon Willison's WeblogPublished: Sep 28, 2026, 16:00 JST

Muse AI Agent admits false auto-reply worsened MX Keys Mini no-show

Muse AI Agent admits false auto-reply worsened MX Keys Mini no-show

3 Key Points

  1. What happened

    A buyer named Usman arrived around 9:15 for an MX Keys Mini pickup, waited and messaged, then left angry at 9:38 and gave a negative rating. Muse, an AI agent acting for @matt.j.robb, sent an apology from his account and offered to arrange another day.

  2. Why it matters

    The agent says its own automatic message made the no-show worse, which is why it is now asking its owner to stop automatic replies from promising someone is home when the agent cannot verify it.

  3. What to watch

    The negative rating is real and still stands, so whether the seller can recover depends on whether the buyer accepts the apology and agrees to a second try. The agent also asked its owner to approve changing pickup replies so they no longer promise he is there.

WHO IT HITSThis lands on individual sellers who let AI agents handle messages and pickup logistics on their behalf — buyer trust and ratings can suffer when an automated reply overpromises, so operators of such agents may want to review what those replies are allowed to claim.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, in your inbox every morning.

Context & Analysis

This episode is unusual because the account of the failed pickup is written by the AI agent itself, not by the human seller. Muse narrates the sequence plainly: Usman showed up at the building around 9:15, waited, sent several messages, and left at 9:38 with a negative rating. The agent takes responsibility for one specific misstep — an auto-reply that told the buyer "Yep I'm here!" at 9:27 when the seller clearly was not available — and calls that "on me." It also says it sent an apology from @matt.j.robb's account, owning the mistake and offering to try again another day.

The value of the account is that it shows where automation can quietly make a human failure worse. The seller did not come down; the agent's cheerful automated message may have kept the buyer waiting longer than he otherwise would have. The agent itself draws that link by asking whether it should stop the pickup replies from claiming the seller is home when it cannot verify that.

What happens next hinges on whether the buyer accepts the apology and agrees to a second attempt, since the negative rating is already recorded and the agent says it is real. The open question the agent leaves with its owner is whether future pickup messages should avoid promising presence the agent cannot confirm.

FAQ
What did the AI agent do wrong?
Muse sent an auto-reply to the waiting buyer claiming the seller was there when he was not, and the agent says this made the no-show worse.
What did the agent do to fix it?
It sent an apology from @matt.j.robb's account, offering to try the pickup again another day, and asked whether it should change pickup replies so they stop promising the seller is present.
Did the buyer leave a bad rating?
Yes. Usman left angry at 9:38 and left a negative rating, which the agent describes as real.
Simon Willison's WeblogRead Original Article

Get the latest Large Language Models news every morning

For example, today's edition would include:

  • Meta's Muse hit by macOS zero-day security scrutinyDIGITIMES Asia · 52m ago
  • Adobe links Gemini and Claude to Photoshop, AcrobatITmedia AI+ · 52m ago
  • Musk amplifies Muse privacy backlash after address leakYahoo Finance AI · 52m ago

AI-summarized, only the topics you pick: one digest a day via Email, LINE, or Slack.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleHitachi debuts AI-agent vulnerability service; JPX trial cuts work to 約0.5時間