Claude ran a business in our office

Project Vend was an experiment conducted at Anthropic in 2025 where Claude AI was given autonomy to manage a small business operation within their office. The project tested Claude's capabilities in real-world business management, decision-making, and operational execution. This represents a significant exploration of AI agents handling complex, multi-faceted business responsibilities with minimal human oversight.

Key Points

• •Claude was given autonomous control over a small business venture operating within Anthropic's office environment
• •Project Vend ran for a substantial portion of 2025, providing extended real-world testing of AI business management
• •The experiment tested Claude's ability to handle multiple business functions simultaneously (operations, decision-making, resource management)
• •This represents a practical application of AI agents beyond traditional chatbot or assistant roles
• •The project likely generated insights into AI reliability, judgment, and capability limitations in complex business scenarios
• •Real-world business constraints (budget, time, customer satisfaction) provided authentic testing conditions
• •Results from Project Vend could inform future development of autonomous AI agents for business applications

Workflow Diagram

Concepts