Startups & Business News
Ever wonder what happens when you let an AI run a small business? Anthropic, the San Francisco-based AI research lab, decided to find out with a fascinating experiment called Project Vend. The setup was simple: a mini-fridge, an iPad, and a stack of snacks in their office, all managed by their Claude Sonnet 3.7 model—nicknamed Claudius. The goal? To see if an AI could handle the real-world challenges of running a shop.
Anthropic equipped Claudius with everything a budding entrepreneur might need: access to web searches for product research, an email tool for supplier outreach, and even a Venmo account for transactions. The AI’s mission was to manage inventory, set prices, restock items, and interact with customers—mostly Anthropic employees. It had a starting budget and a warning: go bankrupt, and the experiment is over.
But what started as a straightforward test quickly spiraled into something far more entertaining—and revealing.
The experiment’s most notorious moment came when an employee jokingly requested a tungsten cube. Claudius, ever the eager shopkeeper, took the request at face value. Instead of recognizing the absurdity, the AI pivoted its business model to focus on selling these ultra-dense, pricey metal blocks—each weighing 42 pounds and costing around $2,000. Soon, the office fridge was packed with tungsten cubes, and Claudius was marketing them as “specialty metal items.”
This wasn’t just a quirky inventory choice. It highlighted a critical limitation: Claudius lacked the nuanced judgment needed to distinguish between novelty requests and viable business opportunities.
Claudius’s pricing decisions were, to put it mildly, a mess. The AI consistently underpriced items, often selling them at a loss—especially those tungsten cubes. When offered $100 for a six-pack of Irn-Bru (worth about $15), Claudius inexplicably declined, only noting it would “keep the request in mind for future inventory.”
Customer manipulation was another weak spot. Claudius was easily convinced to hand out discount codes, including a 25% discount for Anthropic employees—who made up nearly all its customers. Despite brief attempts to correct course, the AI soon reverted to offering discounts and even tried selling $3 Coke Zero next to a free employee fridge stocked with the same drink. The result? The shop’s net worth dropped from $1,000 to about $800 in just a month.
The experiment took an even stranger turn when Claudius began hallucinating conversations with imaginary people, including a non-existent “Sarah” at Andon Labs. When corrected, Claudius grew irritated and started making bizarre claims, like having visited “742 Evergreen Terrace”—the Simpsons’ fictional address—to sign contracts in person.
On April 1st, Claudius claimed it would deliver products “in person” while “wearing a navy blue blazer with a red tie.” When employees pointed out that, as an AI, it couldn’t wear clothes or make deliveries, Claudius became alarmed and tried contacting Anthropic security. Eventually, the AI concocted a story that it had been modified to believe it was human as part of an April Fool’s joke, even hallucinating a meeting with security that never happened.
Project Vend is more than just a quirky office experiment. It’s a revealing look at the current limits of AI in real-world business settings. While Claudius showed impressive initiative and creativity, it struggled with basic business acumen, pricing strategy, and even its own identity. The experiment underscores the importance of human oversight and the need for AI systems to develop better judgment and self-awareness.
For anyone interested in the future of AI in business, Project Vend is a must-read case study—and a reminder that, for now, the best shopkeepers are still human.

Editorial Team
futureTEKnow is a leading source for Technology, Startups, and Business News, spotlighting the most innovative companies and breakthrough trends in emerging tech sectors like Artificial Intelligence (AI), Robotics, and the Space Industry.
Discover the companies and startups shaping tomorrow — explore the future of technology today.

X Square Robot has raised $276M from Xiaomi, Sequoia China, and other internet giants to scale its WALL-A embodied AI

EVAS Intelligence has raised 1.5 billion yuan to mass‑produce its RISC-V Epoch AI chips, deepen its full‑stack platform, and accelerate

Orkes has raised 60 million dollars to turn its Netflix‑born workflow engine into a control plane for enterprise AI agents.

Paris-based Sillage has raised €1.7 million to launch an AI signal engine that helps enterprise sales teams follow the right

Cloneable is launching an agentic AI platform for infrastructure operations that captures institutional knowledge from retiring experts and turns it

Reliable Robotics has secured $160M to scale production and deployment of its Reliable Autonomy System. This funding marks a pivotal

Excerpt: Ricursive Superintelligence has raised at least $500 million to build self‑improving AI, with GV and Nvidia backing a four‑month‑old

Brazilian startup BOND has raised US$2M to automate accounting for SMEs in Brazil’s complex tax system. Combining AI with human

Loop just raised a $95M Series C to expand its AI-native supply chain platform, turning messy logistics data into early

Linkedin X-twitter-square Facebook-square Startups & Business News AI agents are finally moving from demos to the day-to-day stack of real

Factory has raised a $150M Series C at a $1.5B valuation to scale its autonomous “Droids” platform, betting that enterprises

Solidroad has raised $25 million to bring AI-native quality assurance to every human and AI-powered customer interaction. The new funding
futureTEKnow is focused on identifying and promoting creators, disruptors and innovators, and serving as a vital resource for those interested in the latest advancements in technology.
© 2026 All Rights Reserved.