Why Google CEO Sundar Pichai thinks asking 'how much code AI writes' is the wrong question

Google CEO Sundar Pichai has said that measuring the percentage of code written by artificial intelligence (AI) is no longer a meaningful way to assess the technology's role in software development, as AI increasingly influences nearly every line of code shipped by the company.

Pichai previously said that AI generated around 25% of Google's code, a figure that subsequently rose to more than 30%. However, in a recent interview, he indicated that the metric no longer captures the extent to which AI has become integrated into the company's engineering processes.

Instead of focusing on a single percentage, Google is now assessing how AI tools are changing engineers' productivity, the scale of projects they can undertake and the quality of the code they produce.

“What we are tracking with Argon, for example, we are now able to see it’s been a point of inflection. People are using Argon to really go tackle complex projects, including projects which affect millions of lines of code,” he said.

Google's Argon coding agent boosts code submissions

Pichai highlighted Argon, Google's internally deployed coding agent, as a better indicator of AI's growing influence on software development.

“So I think people are writing code more productively and better code with it,” Pichai said. “Nobody would want to code the way they were coding three years ago or even the same time last year.”

According to the CEO, the tool is enabling engineers to move beyond basic code completion and use AI to tackle more ambitious engineering tasks. The shift reflects a broader transformation in how developers interact with AI, with the technology increasingly supporting complex work rather than simply suggesting snippets of code.

Pichai cited improvements recorded during the first four weeks following Argon's rollout as evidence of its impact on engineering productivity.

"In the four weeks since we have launched Argon, we have seen a 30% increase in CLs, and more importantly, almost a similar increase — one third decrease in rollbacks," he said.

CLs, or changelists, are Google's internal term for code submissions. Rollbacks refer to instances in which previously submitted changes have to be reversed.

The figures suggest that engineers are submitting more code while fewer changes are being rolled back, according to Pichai's assessment.

Higher productivity without compromising code quality

The combination of increased code submissions and fewer rollbacks is significant because higher software output does not necessarily translate into better productivity if it also creates additional errors or requires extensive rework.

Pichai's remarks indicate that Google is examining both the volume of work completed and the reliability of the resulting changes to evaluate AI's effectiveness.

However, the figures cited by the CEO relate to the period following Argon's rollout. They do not, by themselves, establish how much of the improvement can be attributed exclusively to AI or whether the gains will persist over a longer period.

AI becomes integral to Google's software development

The comments mark a shift in Google's approach to describing AI-assisted coding. Over the past two years, the company has used estimates of AI-generated code, initially around 25% and later above 30%, to illustrate the technology's growing adoption.

Pichai's latest remarks suggest that this measure has become less useful as AI spreads across the development process.

The technology is increasingly being used for tasks ranging from routine fixes to large-scale code migrations involving millions of lines of code.

As AI becomes more deeply embedded in engineering workflows, Google appears to be placing greater emphasis on measurable outcomes, including productivity, code reliability and the ability to execute complex projects. This approach reflects a broader shift from measuring how much code AI produces to evaluating how effectively it helps software engineers deliver results.