Google says that 75% of its new code already comes from machines

What if much of the software we use every day was already beginning to be written in a different way? AI has been entering programming for some time through the door of the assistants, code suggestions and small automations, but what is beginning to be seen now goes much further. The question is no longer just whether these systems help to write faster, but what happens when a large technology company decides to rely on them systematically. Google has given a pretty clear clue as to where that transition is going.

Google’s jump. The figure was put on the table by Sundar Pichai in a blog post linked to Cloud Next 2026. According to Google’s CEO, the company has been using AI to generate code internally for some time and today 75% of all new code is already generated by AI and approved by engineers. The jump is not minor: last fall, that percentage was 50%. In just a few months, Google has gone from already very high usage to placing AI at the center of much of its software production.

Precision matters. That nuance is not minor: generated by AI does not mean accepted without human control. Pichai talks about code generated by these systems, but also approved by engineers, a necessary difference to not oversize the data. Richard Seroter, Senior Director, Google Cloud, He explained it to Fast Company noting that that human approval is “fundamental in this area.” Google’s reading is that AI can take on an increasing part of production, but within a flow in which engineers continue to validate, correct and make decisions.

Sundar Pichai, CEO of Google

Google’s internal turn. Pichai did not present this advance as a simple productivity improvement, but as part of a shift towards “truly agentive” workflows. As he explained, Google engineers are orchestrating autonomous digital teams, launching agents to complete tasks that previously depended much more on direct human work. The example he cited helps measure the scope of that transition: A complex code migration, performed by agents and engineers, was completed six times faster than was possible just a year ago with engineers working alone.

The engineer changes places. Google’s thesis is not that the programmer disappears, but that their work is displaced. Seroter explained to Fast Company that, with this new distribution of tasks, engineers can focus on higher-value tasks: systems architecture, design and solving complex problems. In this new distribution, manual code writing loses part of its weight and the ability to direct, review and convert those pieces into real products gains importance.

The contrast with the rest of the sector. A Sonar survey from earlier this year notes that 96% of developers acknowledge that they do not fully trust AI-generated code, and that 52% do not always review it for errors before incorporating it. At the same time, the weight of these tools is growing very quickly: the code generated by AI would have gone from 6% in 2023 to 42% in the latest report, with a forecast of 65% for 2027. So we have reasons to say that adoption is ahead of trust.

Images | Xataka with Grok | Stanford Graduate School of Business

In Xataka | A young man has solved a mathematical problem that lasted 60 years in 80 minutes with ChatGPT. That’s the least interesting thing about the story.