How Do AI Agents Spend Your Money? Analyzing and Predicting Token Consumption in Agentic Coding Tasks
The study analyzed token consumption in agentic coding tasks across eight frontier LLMs to understand where agents spend money. Findings show agentic tasks are 1000x more expensive than reasoning, usage is highly variable, and accuracy peaks at intermediate costs. Agents can only predict coarse cost trends themselves, suggesting that while upfront pricing is hard, self-prediction can signal high-cost tasks for better budget alerts.