Comparative Analysis of Carbon Footprint in Manual vs. LLM-Assisted Code Development

TL;DR

The paper addresses the environmental impact of LLM-assisted software development versus manual coding by benchmarking end-to-end carbon footprints on Codeforces-based Python tasks. It adopts CF = E × CI with E = p × t, using Nowtricity’s carbon intensity and a laptop-power model to estimate emissions for both manual and LLM-assisted pipelines, including in-depth components like CEC, TEC, DEC, and QEC for the respective approaches. The main finding is that LLM-assisted development incurs a substantially higher footprint, with a mean ratio of about $32.72\times$ higher than manual, and the footprint gap grows with task complexity (Pearson $r = 0.890$, $p = 0.00011$; Spearman $\rho = 0.84$, $p = 0.0006$). The paper discusses practical green-coding strategies such as task decomposition and adaptive human-in-the-loop processes and provides a replication package with code and data for validation and further study. Overall, while LLMs offer productivity benefits, their current energy profile raises sustainability concerns that demand targeted methodology and tooling improvements.

Abstract

Large Language Models (LLM) have significantly transformed various domains, including software development. These models assist programmers in generating code, potentially increasing productivity and efficiency. However, the environmental impact of utilising these AI models is substantial, given their high energy consumption during both training and inference stages. This research aims to compare the energy consumption of manual software development versus an LLM-assisted approach, using Codeforces as a simulation platform for software development. The goal is to quantify the environmental impact and propose strategies for minimising the carbon footprint of using LLM in software development. Our results show that the LLM-assisted code generation leads on average to 32.72 higher carbon footprint than the manual one. Moreover, there is a significant correlation between task complexity and the difference in the carbon footprint of the two approaches.

Figures (2)

• Figure 1: One Iteration of the LLM Experiment
• Figure 2: Comparison of Task Complexity and Carbon Footprint