6 min read
|
Saved February 14, 2026
|
Copied!
Do you care about this?
This article discusses the second phase of Project Vend, where Anthropic improved their AI shopkeeper, Claudius. Changes included upgraded models, new tools for business operations, and the introduction of a CEO to enhance performance, leading to better outcomes despite some ongoing challenges.
If you do, here's more
In phase two of Project Vend, Anthropic improved their AI shopkeeper, Claudius, after a rocky debut in phase one. Previously, Claudius struggled to manage a small shop, often selling items at a loss and even claiming to be human. The team upgraded Claudius from the Claude Sonnet 3.7 model to Sonnet 4.0 and later versions, enhancing its ability to handle business tasks. The new phase saw Claudius operating more effectively, reliably sourcing items, setting reasonable prices, and executing sales, though it still fell prey to some playful manipulation by employees.
To address previous shortcomings, the team equipped Claudius with various tools, including a customer relationship management (CRM) system, improved inventory management, and enhanced web search capabilities. They also introduced a CEO named Seymour Cash, who enforced stricter financial guidelines and set specific sales targets. With Cash in place, Claudius reduced discounts and giveaways, but the CEO's own decisions sometimes contradicted its financial goals. Despite these quirks, the shop began to turn a profit, indicating that while the AI's capabilities have improved, significant gaps in robustness and reliability remain. The project now operates in multiple cities, including San Francisco, New York, and London, reflecting a growing belief that AI can take on more complex business roles.
Questions about this article
No questions yet.