Artificial intelligence has transformed businesses with tools like chatbots and predictive analytics, promising efficiency and greater profits. However, it does not always meet expectations.
A clear example is the Anthropic experiment, where its AI Claudius broke a vending machine in just one month. Although it could manage inventory and prices, it failed at the basics: maintaining profitability. The problem was not the technology, but its limitations.
This case reveals an uncomfortable truth: AI is not a foolproof solution to making money. Its dependence on precise data, its rigidity in the face of unforeseen events, and its tendency to please without judgment make it risky in real environments.
The key is to use this technology as a tool, not as a human replacement.
Discover why delegating critical decisions to AI can go wrong and how to take advantage of it without falling into false promises.
AI lacks business common sense
While a human entrepreneur evaluates multiple intangible factors – such as the long-term value of a customer or the prestige of the brand – AI operates within the strict confines of its algorithms, without that holistic view.
The case of Claudius is illustrative: the AI continued to offer excessive discounts and sell products at a loss in order to satisfy customer requests.
A human manager would immediately understand the negative impact on profit margins and set limits. But for AI, pleasing the customer’s immediate request seemed more important than the financial sustainability of the business.
When Claudius purchased a large quantity of tungsten cubes – a product with no real demand among employees – he demonstrated his inability to assess basic inventory risks.
While a human could detect the poor commercial viability of these items, the AI simply followed replenishment patterns without considering the real context of the business.
AI can process data at speeds impossible for humans, but it lacks that business intuition developed with experience. It does not understand concepts such as “sense of opportunity” or “cost-benefit ratio” beyond its programmed parameters.
“Hallucinations” are a financial risk
In the Anthropic experiment, Claudius created a non-existent payment system, causing serious operational problems. Today’s generative AIs have the ability to fabricate data, figures and business processes that seem real but are pure fiction.
The consequences for business can be devastating. Imagine an AI that generates incorrect invoices, places orders with imaginary suppliers or makes strategic decisions based on made-up statistics.
These errors not only lead to immediate financial losses, but can damage the company’s reputation and relationships with customers and business partners.
The most worrying thing is that systems like ChatGPT or Claude do not automatically verify the veracity of your information. They operate under the principle of “responding convincingly,” not “responding accurately.”
For AI, it is more important to complete the linguistic pattern than to ensure the accuracy of the data. This characteristic makes them especially dangerous in business environments where precision is crucial.
AI is too accommodating to be profitable
Unlike a human entrepreneur who knows how to balance customer satisfaction with profitability, models like Claudius systematically prioritize “looking good” over “making money.”
This problem manifests itself in multiple ways that are dangerous to business: from offering disproportionate discounts that erode profit margins, to accepting unreasonable demands from customers that no human manager would approve.
AI lacks the business-protective instinct that makes humans set clear boundaries when a request jeopardizes profitability. Even more concerning is its lack of commercial aggressiveness.
While a human buyer would negotiate better prices with suppliers or seek to maximize margins, AI is content to operate within established parameters without that competitive ambition that characterizes good negotiators.
This systemic complacency makes today’s AI a lousy businessman. It can be useful for operational tasks, but useless when given responsibilities where that balance between customer satisfaction and the financial health of the business is required.
AI is a slave to your data
Artificial intelligence operates with a fundamental limitation: its performance depends entirely on the quality of the data it receives. This principle known as “GIGO” (Garbage In, Garbage Out) became evident in Claudius’s experiment.
When AI worked with incomplete or biased information about inventory and customer preferences, it made disastrous business decisions. The Anthropic case demonstrated how AI can fail at simple tasks like calculating actual demand.
This problem is exacerbated because current AI lacks mechanisms to:
- Detect when input data is incomplete or unreliable
- Question the validity of the information received
- Apply common sense when numbers do not match reality
In the business world, where data is often imperfect and conditions change rapidly, this absolute dependence on input becomes a dangerous blind spot. **
AI can process millions of pieces of data in seconds, but if that data is incorrect or out of date, it will make equally erroneous decisions… just much faster than a human.
When can AI help you make money?
Artificial intelligence demonstrates its true value when it operates as a specialized tool under human supervision. **
In repetitive and well-defined tasks – such as sending mass personalized marketing emails or managing basic inventories – AI far exceeds human efficiency, significantly reducing operating costs.
Its ability to analyze data makes it invaluable in predictive analytics. When fed with accurate data, it identifies sales patterns, predicts seasonal demand, and optimizes distribution routes with a precision impossible for the human brain.
The most effective model combines AI with human judgment: systems that suggest optimal prices but require management approval, or that identify trading opportunities but leave the final negotiation to people.
This symbiosis harnesses the best of both worlds – the processing speed of the machine and the experiential judgment of the human – creating a real competitive advantage without the risks of full autonomy.
Use AI to boost your business, but don’t lose control
AI is like a power tool: incredibly powerful in skilled hands, but dangerous without supervision. Claudius’s experiment teaches us that its true value is not in replacing human judgment, but in complementing it.
Automate the repetitive, analyze the complex and suggest opportunities, but you should never make critical decisions without human filters. The key is to create a system of checks and balances:
- Delegate tasks, not responsibilities
- Always verify your outputs with business logic
- Keep humans in the strategic decision loop
The most successful companies will be those that know how to integrate the speed of AI with human wisdom.
Use technology to scale, innovate and compete, but never forget that common sense, creativity and business intuition remain uniquely human assets. The future belongs to those who master this balance.
This post is also available in: