We've officially entered the era of "token engineering." Aka token-efficiency maxxing. The smartest way to cut your AI spend by 50%+ or more: A barbell strategy. First 10% and planning the project: Highest intelligence (Opus/GPT) - this is the most important part Middle 80% (grunt work): Cheaper, open-sourced model for pure execution Final 10% and verification: Switch back to the high-intelligence model I've been doing this for months, and it's hands-down the best way to reduce ridiculous AI spend.(Miles Deutscher)