Consuming too much AI can be a bad idea. New data on 'tokenmaxxing' reveals a better approach. (www.businessinsider.com)

A recent study by Jellyfish highlights the diminishing returns of excessive AI token consumption, revealing that the top 10% of Claude Code users consume about ten times more AI tokens than the median developer yet only double their output. This phenomenon, termed "tokenmaxxing," indicates that spending excessively on AI does not correlate directly with productivity gains, calling into question the sustainability of such practices. Nicholas Arcolano, Jellyfish's head of AI, emphasizes that as companies tighten their budgets, demonstrating effective usage of AI tokens becomes crucial. The findings suggest that the tech industry is leaning towards AI spending discipline, prioritizing efficiency over sheer volume of usage. Jellyfish's data shows that while AI adoption can boost productivity—high-adoption teams posting 77% more pull requests—relying solely on raw token consumption as a productivity metric is misleading. Instead, companies should focus on cost-per-output metrics like pull requests to ensure a balanced approach to AI utilization. Arcolano advocates for broader, moderate adoption of AI tools among developers to maximize productivity without overspending, reinforcing the idea that moderation is key to leveraging AI as a strategic advantage.