- Cursor reports that Nvidia engineers are now committing three times more code than before
- Nvidia maintains that the defect rate remained flat despite the reported increase in output
- AI-assisted workflows contributed to DLSS 4 and smaller GPU die sizes
Nvidia has rolled out generative AI tools across much of its engineering workforce, with Cursor integrated into daily development workflows.
The company says more than 30,000 engineers now rely on this setup, with internal claims pointing to three times higher code output than previous processes.
This claim has attracted attention, in part because volume-based metrics have long been treated cautiously in software development.
Productivity requirements versus engineering reality
This implementation is an operational change that affects core software, including GPU drivers and infrastructure code that supports games, data centers, and AI training systems.
These products are widely considered to be mission critical, where failures can have visible and sometimes costly consequences.
Nvidia claims that the number of defects has remained unchanged despite the increase in output, suggesting that internal controls and testing requirements remain in place.
Driver code, firmware, and low-level system components typically undergo extensive validation before release, regardless of how quickly they are written.
This approach is not new, as Nvidia has previously relied on AI-assisted workflows, including internal systems used to improve DLSS over multiple hardware generations.
Some of Nvidia’s recent achievements are cited as examples of AI-assisted development delivering tangible results.
DLSS 4 and reductions in GPU die size over comparable designs are often cited as results tied to wider use of internal optimization tools.
These examples suggest that AI assistance, when applied in tightly controlled environments, can contribute to measurable improvements.
At the same time, Nvidia’s software stack has faced criticism in recent years, with users pointing to driver regressions and update-related issues across the industry.
Cursor also claims that coding is “a lot more fun than it used to be,” but this ties into productivity numbers that remain difficult to assess independently.
Lines of code committed over a given period of time have never been a reliable indicator of software quality, stability, or long-term value.
True software quality is better measured by stability, maintainability and impact on end-user performance, and output volume alone says little about these.
Nvidia also benefits commercially from promoting AI-powered development, given its central role in providing the hardware behind these systems.
In that context, skepticism about messaging and metrics is to be expected, even if the underlying tools deliver real efficiency gains in specific, tightly managed scenarios.
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews and opinions in your feeds. Be sure to click the Follow button!
And of course you can too follow TechRadar on TikTok for news, reviews, video unboxings, and get regular updates from us on WhatsApp also.
