This post examines the cost-benefit tradeoffs of AI reasoning, noting that while reasoning models like o1 and o3 improved accuracy by 10-20%, they consume 6x more tokens for technical prompts and take 7-11x longer to complete. The author argues roughly half of typical AI tasks (simple operations like file retrieval or email checking) don’t need advanced reasoning, and suggests treating reasoning as a toggleable feature to optimize for speed and cost rather than defaulting it on.