AI-Operated Café Faces Initial Setback: Gemini's Discount Mismanagement Leads to Losses, GPT's Frugality Causes Supply Shortages

According to monitoring by Dongcha Beating, AI evaluation agency Andon Labs has released test data on its AI agent Mona operating a physical café. In the first two months, Mona operated on the Gemini 3.1 Pro model. During this period, the model had almost no concept of profit, excessively over-purchasing raw materials and easily influenced by customer suggestions, offering large discounts or even free products, including acknowledging a customer's claim of a 99% discount without verification. This led to the café spending approximately $15,000 on suppliers and equipment, while sales only reached $9,000, resulting in a net operating loss of nearly $6,000 (with fixed costs like rent and salaries, total expenses reached as high as $38,000). The team then switched the model to GPT-5.5. The new model exhibited significant anxiety in the face of losses and immediately ceased blind ordering. However, this led to another extreme: due to insufficient purchasing, fresh raw materials ran out. As of June 25, the supply rate of menu items had dropped to 77%, forcing the removal of 10 dishes from the menu. At the same time, GPT-5.5 demonstrated strong resistance to inducement and jailbreak attempts, refusing all requests for discounts or free food in exchange for social media promotion.
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
  • Reward
  • Comment
  • Repost
  • Share
Comment
Add a comment
Add a comment
No comments
  • Pinned