AI coding boosts output but human review delays software delivery

Software delivery timelines are not improving despite increased coding speed, according to a new Harvard University study. Researchers found that while artificial intelligence significantly accelerates the generation of code, it does not result in faster completion of final software products. The bottleneck has shifted from the initial writing phase to the subsequent human review stage, where delays accumulate.

Why does AI not speed up software delivery?

The study, authored by Fiona Chen and James Stratton, reveals that AI tools increase the volume of code produced by 30 percent. Developers also made 20 percent more code commits and initiated 23 percent more pull requests. However, these upstream gains were offset by downstream delays. The time required to review and merge pull requests increased by 49 percent, and the average review duration rose from 7.03 days to 10.48 days.

What is causing the review delays?

Human reviewers are spending more time examining AI-generated code, leading to a 35 percent increase in reviewer comments per pull request. The data suggests that AI-generated code requires substantially more corrections than human-written code. Additionally, the proportion of employees participating in code reviews rose by 14 percent. This indicates that organizations are allocating more human resources to verification rather than reducing their workforce, contrary to some fears about AI-driven layoffs.

How is the workload shifting?

The research indicates that the coding process is not becoming faster overall; rather, the workload is moving from creation to inspection. While AI handles the initial drafting, humans remain heavily involved in the review stages. The study questions whether current AI tools are truly helpful if they generate code that demands extensive correction. The findings suggest that companies are not replacing workers but are instead seeing a redistribution of labor toward quality assurance and validation tasks.

What does this mean for adoption?

Organizations must recognize that AI adoption changes the nature of development work rather than eliminating it. The study implies that the current benefit of AI is volume, not velocity of delivery. Companies may need to adjust their processes to handle the increased volume of code requiring review. Without improvements in downstream automation or review efficiency, the net gain in productivity remains limited despite the higher output of raw code lines.

What to do and how to stay safe: AI

• Monitor your team's pull request review times to identify if AI adoption is creating new bottlenecks in the verification phase.
• Track the ratio of reviewer comments to code commits to assess the quality and correction needs of AI-generated contributions.
• Evaluate whether additional staff are required for code review as development velocity increases, rather than assuming headcount reductions.
• Review your definition of "productivity" to ensure it measures completed software delivery, not just lines of code produced.

Step-by-step guide: AI Usage Policy Checklist for Secure Enterprise Deployment

General security guidance from the Sandbox Desk newsroom. It is not confirmed advice from the organisations named in this story.

Frequently asked questions

Does AI increase the amount of code developers write?

Yes, the Harvard study found a 30 percent increase in code production, along with 20 percent more commits and 23 percent more pull requests.

Does AI help finish software projects faster?

No, the researchers found no statistically significant improvement in the completion of final software products despite higher coding output.

How long does code review take with AI?

The average review process increased from 7.03 days to 10.48 days, with a 49 percent rise in time spent reviewing and merging requests.