Google Reveals a Quarter of New Code is AI-Generated, Undergoes Engineer Review

In their latest earnings call last night, Google's CEO Sundar Pichai disclosed that over a quarter of the new code added to Google products is now generated by AI.

However, AI-generated code doesn't make it to production without oversight. Engineers responsible for the products review the generated code before deciding whether to integrate it into the final product. This revelation highlights Google's confidence in its AI capabilities, though it's clear that AI's role in coding is seen as complementary rather than fully autonomous at this stage.

Despite the impressive statistic, Google has not disclosed what percentage of the AI-generated code is eventually adopted. If the adoption rate is low, it could imply that Google engineers spend considerable time reviewing and vetting AI-generated code. This suggests that while AI can produce vast amounts of code daily, not all of it meets Google's standards or requires adjustments.

In the earnings report, Pichai painted a vivid picture of Google's full-stack advantage in AI innovation. This includes a robust AI infrastructure with data centers, chips, and a global fiber network; a leading research team; and a suite of products and platforms that reach billions of users worldwide. With substantial funding, cutting-edge technology, and a vast user base, Google is well-positioned to convert AI advancements into profitable products.

Pichai also shared that Google now has seven products or platforms with over two billion users each, all powered by Google's Gemini model. These services span search, Android, Chrome, Google Play, YouTube, and Gmail.

Recently, Google has restructured internal teams, including merging the Gemini team with the highly capable DeepMind division, to accelerate model development and streamline post-training processes.