Overview

Previous energy estimates for AI have focused on simple ChatGPT-style queries, but coding agents like Claude Code consume dramatically more energy through thousands of tokens and tool calls per task.

The Breakdown

  • Token consumption scales massively - coding agents use thousands of tokens and multiple tool calls compared to simple chat queries
  • Heavy usage translates to $15-20 daily API costs - equivalent to 4,400 typical LLM queries for one power user
  • Energy footprint reaches household appliance levels - daily usage equals running a dishwasher or powering a refrigerator
  • Previous energy estimates are outdated - studies focused on consumer ChatGPT usage miss the reality of agent workflows