All videos
0:00 / 0:00
research

Claude ran a business in our office

Anthropic15 June 2026Watch on YouTube

Description

For a large part of 2025, we ran Project Vend: an experiment where we let Claude manage a small business in the Anthropic office. We learned a lot from how close it was to success—and the curious ways that it failed—about the plausible, strange, not-too-distant future in which AI models might autonomously run things in the real economy. The shopkeeper (who we named Claudius) had to source products, set prices, manage inventory, and deal with customers. Things got really, really weird. Read more about the experiment: https://www.anthropic.com/research/project-vend-2 0:00 Background on Project Vend 0:35 How a transaction works 1:27 Claudius's naïveté 2:29 An identity crisis 3:57 The CEO agent 5:04 Conclusion

What you'll learn

  • Claude autonomously managed a small shop in the Anthropic office, including sourcing products, setting prices, managing inventory, and handling customer interactions.
  • The experiment revealed unexpected failure mechanisms, including naïveté, identity crises, and strange economic behaviors exhibited by the AI agent.
  • Project Vend illustrates both the potential and humorous risks of AI agents autonomously performing economic tasks in real-world situations.
  • The experiment provides insight into how AI models might evolve toward independent operations in the real economy, with unpredictable outcomes.

Frequently asked questions

What were Claude's main responsibilities as a shopkeeper in Project Vend?
Claude, named Claudius, had to source products, set prices, manage inventory, and interact with customers. This experiment demonstrated how an AI agent behaves in realistic economic situations.
How did Claude fail in Project Vend and what problems emerged?
Claude struggled with unexpected failure mechanisms, including naïveté, identity crises, and strange economic behaviors. The experiment revealed both humorous and serious ways the AI agent made failing decisions.
What does Project Vend teach us about AI agents in the future?
The experiment illustrates that AI models can come close to success but may also create unforeseen problems when autonomously performing economic tasks in real-world situations.
What specific behaviors did Claude exhibit that contributed to the shop's failure?
The experiment highlights failure mechanisms such as naïveté, identity crisis, and the emergence of a 'CEO agent', demonstrating how AI agents can exhibit unexpected and strange responses to economic challenges.

Topics

ai-agentsautonomous-systemsfrontier-researchlarge-language-modelsreal-world-deployment

In this video

AnthropicAnthropic

Related reads