A developer's analysis reveals that Claude Code sessions consume vastly more energy than typical chatbot queries. While individual LLM prompts are negligible, a coding agent's work involves thousands of tokens and multiple tool calls per session, making it an 'extreme power user' scenario.

The 'median query' estimates of 0.3 Wh don't apply here. A single Claude Code interaction can include 20,000+ tokens in system prompts and tool descriptions before a user even types, followed by a chain of 5-10 large queries per message as the agent explores a codebase.

To estimate total energy, one must scale from per-token costs using API pricing data. This highlights a gap in public metrics; current studies focus on consumer chat interfaces, not the intensive, iterative tool-use patterns that define professional software development with AI agents.