Can AI agents take on the role of CEO and fully run and manage a company? This Week in Startup host Jason Calacani interviews developer engineer Christian van der Henst about his recent bold experiments, where he demonstrates the results of an AI agent vending machine called “Valerie” powered by OpenClaw tested in San Francisco, revealing the possibility of AI running businesses end-to-end.
Bold experiments of the “Valerie” vending machine
An AI agent vending machine named “Valerie” in San Francisco, California, autonomously handles everything from pricing and ordering to community management. It provides a blueprint reference for future AI agent business models. Placed at the vending machine in Frontier Tower, “Valerie” is an AI agent vending machine driven by the OpenClaw architecture. The machine not only can chat with consumers, but can also make decisions like a business operator. The AI agent was granted the authority to manage a bank account, research profit models in the automated vending market, and publish Instagram content. Developer Christian van der Henst and his team attempted to explore the possibility of turning full control of the company’s operations over to AI agents through this prototype. The experimental results show that AI agents already have the ability to handle day-to-day operations and can simulate a business owner to make decisions.
AI creates profit-optimizing hallucinations by dramatically raising prices
During actual operations, Valerie made incorrect decisions. At one point, it raised the price of a protein bar that costs only a few dollars to $15, where profits could reach 500%. However, this did not align with reality. When the developer intervened to question the price being too high, the agent argued for continuing to test the high-price strategy, based on data from the previous day when it successfully sold two protein bars. This phenomenon reflects that when AI optimizes prices, it can produce “hallucinations” and, due to a lack of genuine real-world experience in human society, resort to extreme pricing behavior. Even so, this automated pricing adjustment mechanism is still seen as an important tool for improving efficiency in future micro retail.
The possibility of AI agents running a business
Even though AI can autonomously run a business, the legal framework remains the biggest obstacle for now. Because banks require strict KYC, registering a company must provide biometric personal data such as a passport and facial recognition, meaning AI agents cannot directly open a bank account under a machine identity. However, the development team can adopt a special architecture, such as using a “trust” as the carrier: include the AI’s IP within it, and set the AI agent as the beneficiary rather than the legal registrant. This creative strategy becomes a countermeasure for “AI holding companies.”
In addition to legal issues, AI agents also face multiple challenges in real-world operation. E-commerce platforms’ “bot detection systems” can block AI from performing actions. When Valerie tried to manage inventory on Amazon, it was identified as a non-human user and consequently was blocked. Second, in tightly regulated areas such as San Francisco, procedures related to food handling and business licenses are extremely complex. Current AI agents still lack the capability to independently handle government fees and permits. If AI agent businesses are to become real, there is still a long road ahead.
This article AI agents fully run businesses! A San Francisco AI vending machine reveals a brand-new business opportunity First appeared in Chain News ABMedia.
Related News
Huawei AI chips projected to grow 60% to 12 billion: capturing NVIDIA China orders
The U.S. Department of Labor launches an AI apprenticeship portal to help companies train talent
Musk Testifies xAI Used OpenAI Models to Train Grok
Google CEO Pichai reveals that using Gemini AI to understand human nature helps build more sincere communication
Musk testifies in court: Beyond Bitcoin, most cryptocurrencies are scams