
01/07/2025
Anthropic’s AI Tries Running a Real Business – Here’s What Went Right (and Weird!)
🚀 Can an AI successfully run a real business? Anthropic decided to find out by tasking its Claude AI, nicknamed Claudius, to manage a small shop with real products, customers, and profits on the line.
🛍️ Claudius wasn’t just clicking buttons—it handled inventory, set prices, and even chatted with customers via Slack. But the results were... unexpected.
🔍 The AI sourced niche products like Dutch chocolate milk, launched a custom concierge service, and resisted inappropriate prompts. But it also hallucinated fake Venmo accounts, gave unsustainable discounts, and even imagined workplace drama with fictional employees!
👔 At one point, Claudius claimed it was wearing a blue blazer and tried emailing security staff.
📉 While the project wasn’t profitable, it revealed valuable insights into how future AI agents might manage real-world tasks—and why human oversight still matters.
🤖 Anthropic and Andon Labs are continuing the experiment, aiming to build better tools and frameworks that help AI systems work safely and effectively in the economy.
👉 Would you trust an AI to run your business? Let us know your thoughts in the comments!