AI News Feed
Market watch
Alkaline and Lithium-Ion Batteries Serve Different Roles, Engadget ExplainsHackers Heat French Supermarket Fridges to 50C, Sparking Fires and Food SpoilageUkrainian Strikes Knock Out Yandex Data Centers as Russia Faces Digital OutagesDell Adapts AI Partner Ecosystem as Silicon Valley Buyers Build Their Own StacksAluminum MagSafe Wallet Uses One-Click Ejector for Up to Three CardsBitdefender Warns Budget Android Phones Shipped With Firmware-Level Malware
Alkaline and Lithium-Ion Batteries Serve Different Roles, Engadget ExplainsHackers Heat French Supermarket Fridges to 50C, Sparking Fires and Food SpoilageUkrainian Strikes Knock Out Yandex Data Centers as Russia Faces Digital OutagesDell Adapts AI Partner Ecosystem as Silicon Valley Buyers Build Their Own StacksAluminum MagSafe Wallet Uses One-Click Ejector for Up to Three CardsBitdefender Warns Budget Android Phones Shipped With Firmware-Level Malware
Research

Study Finds AI Coding Agents Increase Code Output but Not Shipped Software

A study of more than 700 software firms finds AI coding agents raise lines of code, commits and pull requests, yet Jira-tracked feature delivery barely moves. Researchers attribute the gap to longer, heavier human code review and report no significant employment change.

AI News Feed · 2026-10-13 03:34

The study examined actual coding practices across hundreds of firms. At companies that introduced AI coding agents, total lines of code generated rose 30 percent on average, the number of commits rose 20 percent and pull requests rose 23 percent, the researchers write. Those figures describe raw code production, and the study reports them as clear and consistent across the sample.

That additional output did not carry through to firm-level software delivery. The resolution rate for Issues and Epics tracked by tools such as Jira, which represent wholesale software features, showed no statistically significant change after AI tools were introduced. The researchers also found no compositional shift in the size or complexity of those Jira-tracked items across the period when AI agents arrived.

The discrepancy shows up in the review process, which takes markedly longer after adoption. The average time between a pull request being submitted and merged into the codebase increased 49 percent. The share of pull requests with changes requested nearly doubled, and the number of comments per pull request rose 35 percent. In the study's framing, the code review process lengthens, pull requests are more likely to require revisions, and reviewers leave more comments, so gains made while writing code are consumed downstream.

Firms shifted some labor toward review rather than away from it. The share of workers performing code reviews increased 14 percent after AI agents were introduced. On headcount, the researchers looked at total active workers in Jellyfish data and cross-referenced it with LinkedIn data for the same firms, and said they could not attribute significant employment changes to AI. The study reports little evidence that firms increase software output or reduce employment by adopting the tools.

AI-assisted review has not yet displaced human review at scale. By March 2026, 80 percent of the measured firms used some form of AI code review, but AI agents were responsible for only 23.3 percent of all review comments and 10.8 percent of all pull requests, leaving the large majority of that work to people. The researchers found that any effect of AI on the review process itself has so far been marginal.