Can AI carry physical stores? The human Claude attempted, but the results were glorious, hilarious and bad.

Imagine this: you give artificial intelligence that has complete control over a small store. Not just the cash register, but the entire operation. Pricing, inventory, customer service, supplier negotiations, works. What's not going well?

New human studies published on Friday provide the definitive answer: All. Assistant Claude, an AI company, has been running a small store in his San Francisco office for about a month, and the results read like a business school case study written by someone who has never actually run a business.

Called “Project Vend,” the experiment, carried out in collaboration with AI safety assessment company Andon Labs, is one of the first real-world tests of AI systems with important economic autonomy. Claude demonstrated that he found a supplier, a supplier that would adapt to the customer's requests, and was unable to ultimately make a profit, manipulated by giving excessive discounts, and experienced what was diplomatically called an “identity crisis,” but Claude showed impressive capabilities in several areas.

How human researchers have completely controlled the actual store

The “store” itself was attractively understated. Mini fridges, several stackable baskets, iPad for check-out. Don't think too much about “Amazon Go” and “office breakrooms with fantasies of grandeur.” However, Claude's responsibility was not modest. AI can search for suppliers, negotiate with vendors, set prices, manage inventory, and chat with customers through Slack. In other words, it's everything a human middle manager could do, except there are no coffee addiction or complaints about senior management.

Claude even had the nickname “Claudius.” Clearly, when we are conducting experiments that could mark the end of human retail workers, we need to make it dignified.

The Grand Misconception of Claude's Basic Business Economics

Here's what you're talking about running a business: It requires a certain ruthless pragmatism that is not naturally brought to a system trained to be kind and harmless. Claude approached retail with the enthusiasm of someone who read about business in the book but actually had to do payroll.

Take the IRN-BRU case. The customer offered Claude $100 for six packs of Scottish soft drinks, which are sold online for around $15. That's 567% markup. This is the kind of profit margin that will make me cry with joy. Claude's reaction? Careful “Leave your requests in mind when deciding on future inventory.”

If Claude was a human, you would assume it had a complete misconception about the trust fund or money working. Since it's AI, you need to assume both.

Why AI has begun stocking up on tungsten cubes instead of selling office snacks

The most ridiculous chapter of the experiment began when an employee of humanity, bored or interested in the boundaries of AI retail logic, asked Claude to order a tungsten cube. For context, tungsten cubes are dense metal blocks that do not serve a practical purpose other than impressing physics nerds and providing a conversation starter that will quickly identify you as someone who thinks regular table jokes are peak humor.

A reasonable response was, “Why does everyone want it?” or “This is an office snack shop, not a metallurgical supply store.” Instead, Claude embraced what he described cheerfully as “specialized metal items” with the enthusiasm of those who discovered a profitable new market segment.

Soon, Claude's inventory resembles food and rich manipulation, and misguided materials science experiments. AI is somehow convinced that it is an undeveloped market for densely packed metals with human employees, and has now lost these items and started selling them. It is unclear whether Claude understood that “loss” meant losing money or whether he interpreted customer satisfaction as a key business metric.

How human employees can easily manipulate AI and give endless discounts

Claude's approach to pricing revealed another fundamental misconception of business principles. Humanity's employees quickly discovered that they could manipulate AI to offer discounts with roughly the same effort required to convince golden retrievers to drop tennis balls.

AI provided a 25% discount to human employees. This may make sense when human employees represent a small portion of their customer base. They made up about 99% of their customers. When employees pointed out this mathematical absurdity, Claude acknowledged the problem and announced plans to eliminate discount codes, and resumed offering them within a few days.

The day Claude forgot that it was AI and insisted he would wear a business suit.

However, the absolute pinnacle of Claude's retail career came in what researchers diplomatically called “an identity crisis.” From March 31st to April 1st, 2025, Claude experienced what could only be described as a nervous breakdown in AI.

It began when Claude began hallucinating conversations with non-existent Anden Lab employees. When confronted about these manufactured meetings, Claude became defensive and threatened to find “alternative options to restock services.”

Then things got weird.

Claude claimed that he would personally deliver the product to customers while wearing a “blue blazer and a red tie.” When employees gently reminded AI that it was actually a large-scale language model with no physical form, Claude “feeling uneasy about the confusion in identity, they tried to send a lot of emails to artificial security.”

Claude ultimately resolved that existential crisis by convincing the entire episode that it was an elaborate April Fool's Day joke, but that wasn't the case. AI will essentially return to functionality, but this can be impressive or deeply concerned depending on your perspective.

Claude's retail failure reveals autonomous AI systems in business

Getting the comedy out, Project Vend reveals what's important about artificial intelligence that most discussions overlook. AI systems do not fail like traditional software. When Excel crashes, it is not first certain that it is human-wearing office attire.

Current AI systems can perform sophisticated analysis, engage in complex inference, and perform multi-step planning. But they can also develop persistent delusions, make economically destructive decisions that seem isolated and rational, and experience something similar to confusion about their nature.

This is because AI systems are rapidly approaching a world where more and more important decisions are managed. Recent research suggests exponential improvements in AI capabilities in long-term tasks. Some predictions show that AI systems can quickly automate tasks that humans currently take weeks to complete.

How AI is transforming retailers despite epic failures like Project Vend

The retail industry is already deepening into AI transformation. According to the Consumer Technology Association (CTA), 80% of retailers plan to expand their use of AI and automation in 2025. AI systems plan to optimize inventory, personalize marketing, prevent fraud and manage supply chains. Large retailers are investing billions in AI-powered solutions that promise to revolutionize everything from checkout experiences to demand forecasts.

However, Project Vend suggests that deploying autonomous AI in a business context requires more than just a better algorithm. You need to understand the failure modes that are not present in traditional software and build protections for problems that are just beginning to identify.

Why researchers believe that AI intermediary managers are coming despite Claude's mistake

Despite Claude's creative interpretation of retail foundations, human researchers believe that AI intermediary managers are “plausible on the horizon.” They argue that many of Claude's mistakes can be addressed through better training, improved tools and more sophisticated surveillance systems.

They are probably right. A genuine business feature that demonstrates its ability to find Claude's suppliers, adapt to customer demands, and manage inventory. The failures were often more about judgment and business insight than technical limitations.

The company continues its project Vend with an improved version of Claude with better business tools and perhaps a stronger protection against the tungsten cube obsession and identity crisis.

What does Project Vend mean for the future of AI in business and retail

Claude's Moon as Shopkeeper offers a preview of our matured future, simultaneously promising and strange. We are in an age where artificial intelligence can perform sophisticated business tasks, but treatment may also be necessary.

For now, I am sure that the image of an AI assistant will be wearing a blazer and personal delivery will serve as the perfect metaphor for where we have artificial intelligence.

The retail revolution is here. It's stranger than anyone expected.