An employee of the fintech startup Slash accidentally burned $81,000 on AI tokens while creating a meme shooter.
Last week, fintech startup Slash, valued at $1.4 billion, faced an unexpected financial lesson on the efficiency of using artificial intelligence. One of the company's executives, Nicolas Briante, who is responsible for strategic verticals, took an internal call to use AI more actively for writing code more than literally.
Within a single day, Briante used Anthropic's Claude model to create a toy shooter game called Brainrot Shooter. The game was filled with meme characters like Skibidi Toilet and Tung Tung Tung Sahur. However, this creative burst came at a high price: over the week, AI token costs reached $81,267.
An Accident Worth Tens of Thousands
Briante himself called the incident a "real accident." According to him, he underestimated how quickly token consumption accumulates during active development. Each request to the model, especially when loading the context of the entire codebase, consumes resources. As a result, the amount grew to tens of thousands of dollars in just one day of work.
The company responded to the incident with humor. In a post on X, startup representatives joked that employees should play the game so it could be written off as marketing expenses. However, according to available data, the finance department has already reclassified the project from an "expense incident" to a strategic initiative. Slash is now reviewing its policy on using AI for writing code.
From Incident to Strategy: Unexpected Success
Despite the high cost, the game found its audience. In the first 48 hours, 6,912 people played it, with a total playtime of 8,986 hours and an average session length of 1.3 hours. The peak number of concurrent players reached 437. Moreover, the company received three incoming requests for advertising placements.
Slash is far from the only company to encounter this issue in practice. Earlier this year, Uber exhausted its annual AI budget in just four months, after which it introduced its own limits. Another unnamed company received a $500 million bill for one month of using Anthropic's Claude — management simply had not set spending limits for employees.
Expert opinion: This case is a vivid illustration of how the lack of clear limits on AI usage can lead to serious financial losses. Companies need to implement monitoring and API spending limit systems before allowing employees to experiment freely. The irony is that the costly mistake turned into a marketing success and a valuable lesson — but not every startup can afford such a "lesson" for $81,000.