--- title: "Artificial Intelligence Coding Assistants Do Not Necessarily Boost Software Output" url: https://noti.group/artificial-intelligence-coding-assistants-do-not-necessarily-boost/ language: en publisher: "Noti Group" section: "Technology" published: 2026-10-09T19:43:50.000Z updated: 2026-10-10T04:30:47.483Z id: 1cabe0b8-27d2-4a91-80be-a7a62e74612d source: "Ars Technica https://arstechnica.com/ai/2026/10/ai-coding-agents-generate-more-code-but-not-more-software/" attribution: "Link to https://noti.group/artificial-intelligence-coding-assistants-do-not-necessarily-boost/ and name Noti Group when you quote or summarize this story." --- # Artificial Intelligence Coding Assistants Do Not Necessarily Boost Software Output A new study has highlighted a paradox in the use of artificial intelligence coding assistants and agents: while these tools can generate vast amounts of functional code, they do not necessarily translate to increased software output or reduced employment. The research, conducted by Harvard University researchers Fiona Chen and James Stratton, analyzed aggregated data from Jellyfish, which tracks the work processes of engineering teams. This dataset includes over 300 million individual "work events - such as commits and pull requests - and issue management software data from more than 700,000 employees at over 700 relevant software development firms. The study found that human code review forms a significant bottleneck in the efficiency of AI coding tools. Despite the speed with which these agents can generate code, substantial effort is still required to review their output for accuracy. This means that while AI may be able to produce more lines of code, it does not necessarily follow that software output will increase or employment levels decrease. In fact, the study suggests that any efficiency gains from using AI are often offset by downstream constraints in the production process. The researchers used data collected between 2021 and March 2026 to reach their conclusions, highlighting a trend that may have significant implications for the way companies approach software development in the future. The study's analysis of GitHub activity revealed a significant increase in code production following the introduction of AI coding agents at various companies. This surge in productivity was quantified by a 30 percent rise in total lines of code generated, accompanied by a 20 percent increase in commits and a 23 percent uptick in pull requests on average. However, despite this substantial increase in raw code output, the quality of the software produced did not see a corresponding improvement. The resolution rate for Issues and Epics tracked by tools like Jira remained unchanged after AI tools were introduced, indicating that the additional code was not translating into functional software features. Moreover, the researchers found no evidence of a compositional shift in the size or complexity of issues being addressed across firms that adopted AI coding agents. This suggests that the increased productivity may be attributed to the efficiency and speed at which code is generated rather than any actual improvement in software quality. The introduction of AI tools has sparked debate about their potential impact on the software development process. While some argue that these tools can streamline workflows and enhance productivity, others raise concerns about their ability to generate high-quality software features. As the study's findings continue to shed light on the effects of AI coding agents, it remains to be seen whether companies will need to reevaluate their approach to software development in response to these emerging trends. The average time it takes to review code after introducing AI coding agents has increased significantly. According to researchers, the review process" time balloons by 49 percent on average following the introduction of these agents. This delay is not just a minor blip in the workflow, but rather a substantial increase that affects the entire development process. The researchers have dug deeper into this issue and found some surprising statistics. After AI agents are introduced, the share of pull requests with changes requested nearly doubles, and the number of comments per pull request increases by 35 percent. This suggests that human reviewers are still playing a crucial role in the code review process, but with added complexity. In response to these changes, researchers observed a slight increase in the number of workers performing code reviews after AI agents' introduction. Specifically, they found a 14 percent rise in the share of workers engaged in this activity. However, when examining total active workers across Jellyfish and cross-referencing with LinkedIn data, the researchers concluded that AI has not had a significant impact on employment levels. The use of AI code review tools is widespread, but their actual contribution to the development process appears limited. By March 2026, a substantial 80 percent of measured firms were using some form of AI code review. However, AI agents accounted for only a small fraction of all review comments and pull requests, with 23.3 percent and 10.8 percent respectively. While AI could potentially streamline the review process, its impact has been minimal so far. The researchers' findings suggest that human reviewers continue to bear the bulk of this work, despite the presence of AI agents in the development pipeline. The integration of AI agents into software development has brought about significant changes in the industry, but their impact on productivity remains a topic of debate. Despite widespread adoption, with 95 percent of firms in a recent study having implemented AI coding agents, many companies are still navigating the optimal use of these tools. As software engineering teams gain experience, they may be able to better balance coding time and review time. The current state of affairs suggests that while AI agents can accelerate coding speed, they also lead to increased human code review time and effort, making it unclear whether their implementation is justified by cost and time savings. --- Source: [Ars Technica](https://arstechnica.com/ai/2026/10/ai-coding-agents-generate-more-code-but-not-more-software/) Published by Noti Group: https://noti.group/artificial-intelligence-coding-assistants-do-not-necessarily-boost/