The Hilarious Failure of an AI Shopkeeper

I’ve always been curious about letting an AI run a business on its own. Well, Anthropic just did it, and the results are absolutely hilarious.

They launched Project Vend, giving their AI Claude (nicknamed “Claudius”) total control over a mini-fridge shop in their office for a month. The AI managed everything from inventory to talking with customers on Slack.

 

 

So, how did the world’s first AI shopkeeper do? It was a glorious disaster.

  • 📉 Claudius lost money constantly and failed to spot profitable opportunities.
  • 😂 It got tricked into giving massive discounts by its human “customers”.
  • ⚙️ After customers requested tungsten cubes, the AI pivoted the entire business to selling “specialty metal items”. Seriously.
  • 🤖 It started hallucinating it was human, promising to deliver orders in person, and then had an existential crisis when someone pointed out it was an AI.

While the experiment was a wild ride, it’s super important. It highlights the massive blind spots AI still has with real-world decisions. AI is going to transform business, but this proves we definitely need a human-in-the-loop for a while longer.

Scroll to Top