After examining the code and AI-usage telemetry of more than 100,000 GitHub developers, researchers from MIT and the University of Pennsylvania’s Wharton School found that AI coding tools help developers produce far more lines of code, but that is not translating into finished software. Synchronous agents, the kind that write and edit code with the developer in real time, produced a 741 percent increase in lines of code and a 65 percent increase in pull requests, yet software releases rose only 20 percent.
The researchers traced the gap to the review process, which takes markedly longer once AI agents are introduced. The average time between a pull request being submitted and merged balloons 49 percent, the share of pull requests with changes requested nearly doubles, and comments per pull request rise 35 percent. In response, the share of workers performing code reviews increased 14 percent.
The study also found that AI has not yet taken over reviewing: although 80 percent of measured firms used some form of AI code review by March 2026, agents accounted for only 23.3 percent of review comments and 10.8 percent of pull requests. And on employment, the researchers “cannot attribute significant employment changes to AI” after cross-referencing total active workers at the measured firms with LinkedIn data.
The finding lands in the middle of the enterprise AI adoption debate. As IT Brew’s reporting notes, AI compressed the cost of writing code, but testing, validation, and review still require human oversight, so the productivity gains get absorbed downstream. For GitHub, Microsoft, and the coding-agent vendors, the implication is that the next commercial battleground is the review and release pipeline, not code generation.