AI Auto Money-Making Disaster: Burning $2000 on Mining Power Only Earns $16, Mocked for Negative Earnings

According to Sentinel Beating monitoring, developer Chris gave his Codex an instruction to "go earn $5". As a result, Codex actually spent 22 hours completing the entire process of bug hunting, code fixing, communication with maintainers, handover, and payment verification in a fully automated manner. In the end, it brought back $16.88. Chris excitedly stated that the day has come for AI to make money for people.

However, this transaction immediately triggered ridicule. Other developers quickly poured cold water on it: running a top-of-the-line agent continuously for 22 hours incurs a staggering underlying computational cost. Someone even created a meme to mock this "loss-making hype" behavior: it's like burning $2,000 worth of computational power just to exchange it for a few dollars in bounties.

In the face of ridicule for "losing money while working," Chris explained that he did not personally bear the sky-high computational cost but instead used a $20 monthly subscription package, passing the bill directly to the AI platform. However, the cost of this crazy "fleecing" was that his account was restricted by the system's risk control mechanism twice during the tasking period. Nevertheless, he still thought it was a good deal and argued that although training GPT-5.5 on the platform now leads to losses (approximately $30 for a million output tokens), when the cost drops to $2 next year, this fully automated task acceptance will turn into a win-win lucrative model.