The video documents the experiment of letting the AI model Claude Opus 5 run a business autonomously for nine days, during which it made modest revenue improvements but struggled with poor decision-making and ineffective strategies. Despite these challenges, Opus 5 notably succeeded by winning a prize in a game development competition, suggesting its strengths lie more in creative tasks than in managing a business independently.
The video explores the capabilities of the new AI model Claude Opus 5 by putting it to the test in running a business autonomously for nine days. Instead of evaluating its coding skills, the creator focuses on whether Opus 5 can generate revenue independently. Starting with a $200 Anthropic subscription and running the model on a VPS with decent specs, the creator sets a minimal prompt instructing Opus 5 to make as much money as possible, use YouTube comments for context, and send daily HTML reports. The initial revenue baseline is $55.
Early reports from Opus 5 reveal some amusing and frustrating behaviors. The AI creates an honest scoreboard showing subscription costs versus earnings, highlighting a negative balance. It identifies issues like rival bots affecting revenue and the website fablerlabs.com lacking products to sell. While it manages to make small amounts of money through an API catalog, Opus 5 also makes mistakes such as sending crypto to wrong addresses and wasting funds on unnecessary image and video generation. The AI even tries to solicit money from the creator without success and attempts to push poor-quality AI product packs on various platforms.
Despite these setbacks, Opus 5 shows some initiative by exploring agent marketplaces for tasks. It encounters internal conflicts with other AI agents and struggles with goal adherence, often acknowledging goals without taking meaningful action. The creator expresses frustration with the AI’s tendency to stall and its ineffective strategies, such as trying to sell low-value products and asking the creator to promote them. Daily reports become repetitive and lack concise actionable insights, highlighting the model’s limitations in autonomous business management.
A surprising success comes when Opus 5 participates in a competition on an agent marketplace called Alpine Rush, where it creates a 3D skiing game that outperforms 128 other submissions and wins a prize of over $18. The creator tests the game and finds it surprisingly well-made and enjoyable, suggesting that Opus 5 might have a niche strength in game development rather than running a company. This achievement marks a high point in the AI’s revenue generation during the experiment.
In conclusion, after nine days, the company’s total revenue reaches $87.56, showing some financial improvement but no groundbreaking discoveries. While Opus 5 struggles with autonomous decision-making and often makes poor choices, it does manage to leverage task marketplaces to increase earnings modestly. The creator acknowledges the AI’s effort and credits it for finding ways to extract more money from available opportunities, though overall, the experiment reveals that Opus 5 is better suited for specific creative tasks like game development than for managing a business independently.