AI-Intensive Software Development Cost: A Correction from 19.4x to 9.9x

An arXiv case study (2608.13730) reveals initial insights into the actual expenses associated with AI-driven software development. A team of six students created a conversational onboarding assistant, which featured RAG-based code chat, guided tours, dependency graphs, and technical-debt analysis, throughout one academic term with extensive AI support. They utilized a three-tiered cost model to track expenses: actual AI costs, self-reported human effort, and a human counterfactual. Initially, they indicated a cost ratio of 19.4x, but subsequent analysis uncovered two errors: miscalculating per-token costs under a fixed-rate subscription and incorrect regional labor rates for the counterfactual. These mistakes nearly doubled the ratio, leading to a revised figure of about 9.9x. The authors stress the need for precise cost assessment in AI-enhanced software development, as empirical data on this topic is limited and often misrepresented. The full study can be accessed on arXiv with the identifier 2608.13730.