Sarthak Dhawan, a 21-year-old co-founder of Turbo AI, shares a cautionary tale about the unintended consequences of AI token spending. In a single month, his startup spent a staggering $30,000 on Claude Code tokens, a seemingly small detail that highlights the evolving relationship between humans and AI in the workplace.
Dhawan emphasizes that this wasn't a total mistake, but rather a reflection of their ambitious shipping goals. He and his co-founder, Rudy, launched their AI learning tool app in January 2024 while still in college, and later dropped out to pursue their business full-time. Their bottleneck, according to Dhawan, is shipping speed, not token spending.
The author delves into the transformative impact of AI on the coding process. He explains that AI has shifted his job from writing code to reviewing it. He spends more time planning and describing high-level system behavior and scrutinizing AI-generated code than writing his own. This shift has led to concerns about his coding skills atrophying, a phenomenon shared by many engineers using Claude Code.
The article highlights a stark contrast between the past and present of software engineering. Twenty years ago, engineers deeply understood the intricacies of their codebases, having written them from scratch. However, with AI taking over the coding process, the codebase becomes an enigmatic entity that engineers struggle to comprehend. Despite this, Dhawan acknowledges the productivity gains AI brings, making it difficult to resist its use.
Turbo AI's approach to AI token spending is pragmatic. They don't have a strict budget, but rather keep a loose eye on costs. This flexibility allows them to experiment and innovate without being constrained by token limits. The team averages around $20,000 a month on AI tooling costs for software development, and they measure token costs per developer by encouraging usage without strict approval processes.
The story takes a humorous turn when Dhawan reveals that his excessive spending on Claude tokens was due to a simple oversight. He had left the 'fast mode' setting on, which significantly increased token costs. This revelation underscores the importance of vigilance in managing AI settings, even for experienced users.
Dhawan concludes by emphasizing the need for a balanced approach to AI token spending. While AI can significantly enhance productivity, it's crucial to be mindful of costs. He encourages a pragmatic mindset, prioritizing output and long-term business value over every dollar saved. The article leaves readers with a thought-provoking question: How can we strike the right balance between embracing AI's potential and managing its financial impact?
