'A rocket ship.' AI is doubling software output, and code quality is holding up

The software development landscape is undergoing a seismic shift as companies increasingly adopt AI coding tools, leading to significant improvements in output. A comprehensive benchmark study conducted by engineering intelligence platform Jellyfish has revealed that over 700 companies, comprising 200,000 engineers and analyzing 20 million pull requests, are experiencing this transformation. Currently, the adoption of AI tools in software development is at an impressive median rate of 63%. Notably, 64% of the organizations analyzed are now producing the majority of their code with AI assistance. The data from Jellyfish indicates a steady increase in weekly usage, with more engineers utilizing AI coding tools several times a week. The most striking statistic arises from companies employing AI tools extensively, where engineers merge an average of 2.2 pull requests each week. This figure is nearly double the 1.12 pull requests observed in companies with low AI adoption. A pull request, or PR, represents a proposal for code changes in a shared project, requiring review and approval before implementation. Nicholas Arcolano, head of research at Jellyfish, noted that the perception of AI as a tool for accelerating coding is becoming widespread. He also raised important questions regarding the return on investment and quality implications of these advancements. After fully transitioning to AI coding tools in late 2025, Arcolano described a pivotal moment he dubbed 'Claude Christmas', when many engineers discovered the potential of Anthropic's Claude Code service. In addition to Claude, other popular AI coding tools such as OpenAI's Codex and GitHub Copilot are gaining traction among developers. Encouragingly, the study indicates that the quality of code remains stable despite the rapid increase in output. The revert rate, which tracks the percentage of code that needs to be rolled back post-deployment, sees only a slight increase from 0.61% in low-adoption firms to 0.65% in those fully embracing AI. Arcolano expressed optimism about the future, highlighting that while the quality of code has not significantly declined, it will be crucial to monitor this aspect as the pressure for faster delivery mounts. Moreover, the emergence of autonomous agent activity in coding—where AI agents initiate or commit pull requests—is on the rise, especially among leading adopters. The divide between organizations fully leveraging AI and those lagging behind is becoming increasingly pronounced. "The people at the bottom are not moving, the people in the middle are progressing slowly, while those at the top are on a rocket ship, accelerating away with AI," Arcolano concluded, capturing the essence of the current AI adoption narrative.