Real-time visibility into the carbon cost of AI-assisted coding, powered by independent academic research
AI coding tools have reshaped how software gets built. Yet the environmental cost of this shift is almost entirely unmeasured: ten of thirteen major AI companies disclose nothing about the environmental impact of building and deploying their models, according to Stanford’s 2025 Foundation Model Transparency Index.
We built Carbonlog because measurement has to come before management, and right now, almost nobody is measuring.”
— Mark Chen, Co-founder and CEO, CNaught
CNaught is launching Carbonlog, an open-source Claude Code plugin that tracks the carbon emissions and energy consumption of AI-assisted coding sessions in real time. It is the first developer tool to provide per-session emissions data based on independent academic research.
A Growing Footprint with No Visibility
The scale of AI’s environmental impact is becoming clearer. The International Energy Agency projects that global electricity demand from data centers will more than double by 2030, reaching 945 terawatt-hours (TWh), more than Japan’s total electricity consumption today. Goldman Sachs Research estimates that if 60% of new data center power demand is met by natural gas, the resulting emissions increase could reach 215–220 million tonnes of CO₂ by 2030.
Marketing Technology News: MarTech Interview With Fredrik Skantze, CEO and Co-founder of Funnel
Until now, developers haven’t had an easy way to measure their contribution to this footprint. Estimation methodologies exist in the academic literature, but building them into a usable tool has been left as an exercise for the climate-conscious developer. CNaught has taken the best available methodology and made it turnkey: install one plugin, and see your emissions every time you submit a query.
CNaught is committed to using the best available science and keeping the methodology current as the research evolves, so teams don’t have to track the literature themselves. The underlying framework is already being updated as new model benchmarks become available.
What Carbonlog Does
Carbonlog runs as a Claude Code plugin, estimating the energy consumption and CO₂ emissions of each API request in real time. A subtle status line displays cumulative emissions as developers work. An on-demand report provides breakdowns by model, project, and time period, with relatable equivalents like car-miles driven and days of household energy usage.
Marketing Technology News: The Death of Third-Party Cookies Was Just the Start. Are You Ready for Consent Orchestration?
The plugin’s calculations are based on Jegham et al. (2025) “How Hungry is AI? Benchmarking Energy, Water, and Carbon Footprint of LLM Inference.” This framework estimates energy from first principles using real-time performance benchmarks, statistically inferred hardware configurations, and provider-specific infrastructure parameters.
All data is stored locally by default. No code, conversation content, or personal information is shared. Developers can optionally enable anonymous metric sync for team-level benchmarking through CNaught.
Why It Matters Now
California’s SB 253 will require companies with over $1 billion in revenue to report Scope 3 emissions starting in 2027. For organizations using AI inference as a purchased service, those emissions typically fall under Scope 3, Category 1: Purchased Goods and Services. Carbonlog provides the data needed to establish a baseline before reporting requirements take effect.
What Comes Next
Carbonlog is currently in beta with a group of design partners. Installation takes approximately 30 seconds.
Carbonlog launches as a Claude Code plugin, with plans to expand to other AI coding tools and chat-based LLM usage. Beyond measurement, CNaught is exploring how Carbonlog can move from visibility to recommendations on how to build greener software.
