on wednesday, local time, google disclosed that three-quarters of the code currently being written in-house is now generated by ai and then reviewed by human engineers. this proportion has been steadily rising in recent years—back in october 2024, it was about one-quarter, and by last fall it had climbed to 50%. in a blog post, google ceo sundar pichai stated that the company is shifting toward “truly agent-centric workflows,” enabling engineers to take on more autonomous tasks. he cited as an example a recent complex code migration project carried out collaboratively by agents and engineers, which was completed six times faster than a year earlier when it was handled solely by engineers.
google engineers are currently using the gemini model to generate code, and some employees have even been assigned specific ai-use targets that are factored into their annual performance reviews. interestingly, in recent months, certain google deepmind employees have been authorized to use claude code from rival anthropic, a development that has sparked some internal tension.
google is not the only tech giant ramping up its use of ai in programming. in april last year, microsoft ceo satya nadella announced that ai-generated code already accounts for 20% to 30% of some of microsoft’s projects, while the company’s cto predicts that within five years, 95% of all code will be generated by ai. meta is also pushing hard, aiming to have 55% of code changes in certain teams fall under the “agent-assisted” category by the fourth quarter of 2025. earlier this month, snap reported that at least 65% of its new code is now generated by ai. as major companies increasingly embrace ai-assisted programming, the practice is moving from the experimental stage to mainstream adoption, and the role of engineers is accelerating its shift from coders to reviewers.
