How to Actually Measure the Carbon Footprint of AI Code
Summary
Introduction to Carbonlog, a Claude Code plugin that estimates real-time CO2 emissions from AI coding sessions using the Jegham et al. (2025) academic methodology.
Key quotes
Carbonlog is a Claude Code plugin that tracks the carbon emissions of your AI coding sessions in real time, using an independent methodology from leading researchers.
We prioritize independent academic methodology over provider-disclosed figures, which carry inherent conflicts of interest.
Treat results as order-of-magnitude estimates, not exact measurements.
The post details the technical implementation of Carbonlog, explaining its use of TTFT and TPS benchmarks from Artificial Analysis and its categorization of emissions as Scope 3, Category 1 for organizational reporting. It contrasts its methodology with the EcoLogits library, specifically regarding assumptions about batch size and hardware.