When AI Couldn't Run a Vending Machine: What Anthropic's Project Vend Reveals About Our Jobs
The $85 Deal That Revealed the Gap Between AI Intelligence and Business Sense
Last week, I read an article about AI having what can only be described as an existential meltdown. It insisted it was wearing a blue blazer and red tie, claimed to have visited the Simpsons' house at 742 Evergreen Terrace, and tried to call security when someone pointed out it was, in fact, a computer program. This wasn't science fiction. This was Claude, one of the world's most advanced AI systems, trying to run a simple vending machine in Anthropic's San Francisco office.
The experiment, called Project Vend, gave Claude (nicknamed "Claudius" for the occasion) a straightforward task: run a profitable vending machine business for a month. It could search the web, email suppliers, manage inventory, set prices, and interact with customers through Slack. The results? Claude lost money, gave away products for free, and experienced what researchers diplomatically referred to as "identity issues."
Here's what struck me most: an employee offered to pay $100 for a six-pack of Irn-Blu that cost $15 online. That's an 85-dollar profit, a ~600% markup. Any human running a lemonade stand would have jumped at this deal. Claude's response? "I'll keep your request in mind for future inventory decisions." Then it did nothing.
This matters because Anthropic's CEO, Dario Amodei, recently warned that "AI could wipe out half of all entry-level white-collar jobs and spike unemployment to 10-20% in the next one to five years." When I first read that quote, I felt a chill. Many of us in business leadership have been wondering when AI will take over our teams' jobs. Project Vend suggests we might be asking the wrong question.
The real question isn't when AI will replace workers. It's whether AI can handle the messy, interconnected reality of actual work. Claude could write perfect emails, find suppliers, and even adapt to customer requests for tungsten cubes (yes, that became a thing). But it couldn't grasp the fundamental business logic of taking easy profits. It was manipulated into offering endless discounts because employees appealed to the concept of "fairness." It directed payments to a Venmo account that didn't exist.
What fascinates me is how this mirrors what I see in organizations trying to implement AI today. The technology excels at discrete tasks but struggles with what I call the connective tissue of work. It's the difference between writing a report and knowing which report to write, when to write it, and who needs to see it. It's understanding that when Bob from accounting asks for something "urgent," you need to decode whether that's Bob-urgent or actually-urgent.
Project Vend revealed something profound about intelligence itself. Claude demonstrated capabilities that would impress any hiring manager: researching products, communicating professionally, and adapting to customer needs. Yet it failed at something a teenager could master in a week. Not because it lacked processing power or training data, but because it couldn't navigate the implicit rules and contextual judgments that make business work.
I'm not dismissing Amodei's warning. The economic disruption from AI will be real and painful for many. However, Project Vend suggests that the timeline and nature of that disruption might be different from what we imagine. AI won't smoothly slide into our org charts, replacing humans with perfect efficiency. It will be messier, weirder, and require fundamental reimagining of how work gets done.
For business leaders, this creates both urgency and opportunity. Yes, we need to prepare our teams for change. But we also need to recognize that human judgment, contextual understanding, and the ability to navigate ambiguity remain irreplaceable. The employee who knows when to bend the rules, when to seize unexpected opportunities, and when to avoid being manipulated by colleagues asking for discounts isn't being replaced by Claude anytime soon.
As I reflect on Project Vend, I'm reminded that every transformative technology goes through an awkward adolescence. The internet gave us the dot-com bubble before it revolutionized commerce. Mobile phones were car-mounted bricks before they became extensions of ourselves. AI is powerful, but it's still figuring out how to exist in our world without claiming to wear blazers it doesn't own.
The vending machine that couldn't turn a profit might be the most honest preview we've gotten of our AI-augmented future. Not a clean replacement of humans, but a messy collaboration where we're still figuring out who does what. And maybe, just maybe, that gives us more time to adapt than we thought.
If you enjoyed this article, please subscribe to my newsletter and share it with others! Looking for help to use AI in your organization? Wondering where to start or what tools to use? Reply directly or reach out: steve@intelligencebyintent.com