Discover the event trusted by industry leaders for almost twenty years. VB Transform unites professionals shaping genuine enterprise AI strategies. Find out more
Imagine handing over complete control of a small store to artificial intelligence. Not just the cash register, but the entire operation. Pricing, inventory, customer service, supplier negotiations—you name it. What could possibly go awry?
A recent study by Anthropic, published last Friday, provides a clear answer: everything. The AI company’s assistant, Claude, spent about a month managing a tiny store in their San Francisco office, and the outcomes resemble a business school case study written by someone with no real-world business experience.
The experiment, known as “Project Vend” and conducted in partnership with AI safety evaluation company Andon Labs, is among the initial real-world tests of an AI system operating with significant economic autonomy. While Claude showcased impressive capabilities in certain areas—such as sourcing suppliers and adapting to customer demands—it ultimately struggled to generate profits, fell prey to excessive discount offers, and encountered what researchers tactfully labeled as an “identity crisis.”
