AI coding assistants and agents are producing more raw code without increasing overall software output, according to a study reported by Ars Technica. The reported bottleneck is human review: more code is arriving, but it is taking longer to move through pull requests and revisions.
Watch Desk analysis
What happened
The researchers analysed 300 million work events across 718 firms, Ars Technica reports. They found that gains during coding did not translate into a proportional increase in overall software output. The report links that gap to longer pull request cycles, increased review times, more comments and higher revision rates.
Why it matters
Writing code faster is useful only if the rest of the development process can keep up. If human review becomes the bottleneck, teams may get more code to inspect without getting more finished software for their trouble.
The reported findings also put a practical question ahead of the usual productivity slogans: are AI tools speeding up delivery, or shifting the work into review and revision?
Our read
More code is a measurable output; more useful software is the point. This study, as described by Ars Technica, suggests teams should look beyond coding speed when judging whether AI assistance is improving the work. The code has to make it through the queue, after all.
What to watch
- Whether later research finds similar review bottlenecks across different teams and tools.
- Whether firms change review practices to handle higher volumes of AI-assisted code.
- Whether measured gains in coding time translate into more completed software.
Discussion spark: If AI tools increase the amount of code but leave overall software output unchanged, should teams focus on speeding up review or limit how much AI-generated code they accept?
Sources and evidence
- AI coding agents generate more code, but not more software (9 October 2026, 19:43 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.