Software developers across various sectors, notably in finance, are increasingly adopting AI coding tools, with significant implications for productivity and code quality. Investment banking quants have described Microsoft’s CoPilot as a “revelation,” while fintech firms report generating approximately 50% of their code through advanced tools such as Cursor. While these technologies have resulted in impressive metrics of developer productivity, questions loom about potential unforeseen consequences.
In a report released yesterday, Greptile, an AI code review startup, analyzed the impact of AI tools on productivity metrics within its teams, revealing a 76% increase in lines of code (LoC) per developer from March to November 2025. Medium-sized teams, comprising 6 to 15 members, exhibited an even more remarkable increase of 89%. Although the size of pull requests increased by 33%, the number of lines changed within each pull request rose by only 20%. This indicates that while developers are producing significantly more code and submitting more pull requests, they are editing their existing code to a lesser extent.
This shift towards increased code output, however, does not automatically signify improved quality. The report did not evaluate the quality of the code produced, leaving it unclear whether there was a meaningful change in the number of bugs found or fixed. The findings suggest a paradigm shift from traditional code evaluation to a focus on content production, which could lead to complications. Some developers might receive recognition for merely increasing LoC, potentially leading to inflated output favoring quantity over quality. A FAANG engineer commented on Hacker News that they had noticed “an uptick in the number of lines on pull requests,” attributing this to both AI coding tools and performance review incentives.
As the report highlights, viewing lines of code as assets may be misguided; many industry voices argue they should be seen as liabilities. This perspective raises a crucial issue: if productivity metrics similar to those observed in Greptile’s report are evident across other companies, the implications could be substantial. Notably, OpenAI’s software development kit has been downloaded roughly 130 million times since 2022, and Anthropic’s has seen over 30 million downloads. Should all developers leveraging these kits double their output, the potential for hundreds of millions of additional lines of code per year would arise, while the number of developers available for code review remains relatively static. The result could be an increased likelihood of defects going unnoticed, compounded by a potential surge in technical debt stemming from excessive code production that may eventually require extensive rewriting.
Given the heavily regulated nature of banks, which often require permission for even minor changes to a codebase, this industry may introduce bottlenecks into the productivity gains realized by AI tools. Nevertheless, banks have been developing AI solutions aimed at reducing the developer hours spent on tasks like rewriting legacy code. If the time saved is redirected toward analyzing the influx of new code, the outcome could be beneficial. However, if these hours are instead used to justify workforce reductions, the same risks related to code quality may persist.
The rapid progression and integration of AI coding tools suggest that a critical examination of their impact on the software development landscape is warranted. As developers continue to adapt to these technologies, the balance between productivity and quality will be pivotal in shaping the future of code creation and evaluation. The conversation surrounding metrics of success in the era of AI coding tools is likely to evolve, with significant ramifications for both developers and the wider tech industry.
See also
AI Recruiting Software Revolutionizes Hiring: 12 Essential Tools for 2023
London’s AI Software Development Surges, Enhancing Business Efficiency and Innovation
Microblink Enhances IDV Software with Real-Time Liveness Checks to Combat AI Fraud
Gen AI Platforms Drive Research but Show Low Holiday Purchase Intent Among Consumers
Kie.ai Launches Grok Imagine API for Rapid, Cost-Effective Product Video Automation
