A developer details the significant costs associated with using coding agents, particularly those powered by models like Claude Code. The primary cost driver is not the agent's output, but the large context window it rereads on every turn, which can amount to approximately 74,000 tokens. While prompt caching reduces this cost, it is not entirely eliminated, and the expense can escalate rapidly, especially when using subagents. The developer observed a dramatic increase in weighted token usage over several months, correlating with increased delegation to subagents. AI
IMPACT Developers using AI coding agents should be aware of the significant token costs associated with context rereading, which can impact subscription usage and overall expenses.
RANK_REASON The item discusses the operational costs and efficiency of using an AI coding agent, rather than a new model release or research.
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →