75% of New Code at Google Generated by AI
Google has disclosed that approximately 75% of its new code is generated by AI, becoming a major production method in the software development process. This data was revealed during the Cloud Next event, reflecting the deep integration of AI into engineering systems.
The company also stated that its models can currently handle over 16 billion tokens per minute, supporting large-scale code generation and application deployment. English media widely view this change as a sign of AI shifting from an "auxiliary tool" to a "core productivity" driver.
A similar trend has been observed in disclosures from companies like Microsoft and Meta, with several tech giants increasing the proportion of AI-generated code in their development processes.
Source: Public Information
ABAB AI Insight
75% indicates a structural break in the software production function. Code is no longer primarily written line by line by humans, but generated by models, with humans shifting to defining problems, reviewing results, and system design. This essentially transforms "coding" from a skill into an interface, concentrating productivity on fewer high-level decision-making and architectural capabilities.
This will directly change labor pricing. The marginal value of junior and mid-level engineers is compressed, as their work is most easily replaced by models; whereas engineers who can design systems, break down problems, and validate complex outputs will see their value increase. The software industry will experience more pronounced stratification internally, rather than benefiting as a whole.
From a company perspective, this shift increases output elasticity. The cost of code generation approaches that of computing power, allowing software development to scale on demand like computational resources. This will accelerate product iteration speed while lowering trial-and-error costs, reinforcing the scale advantages of leading companies.
In the longer term, this represents a phase in "cognitive automation." In the past, automation replaced manual labor; now it is beginning to systematically replace structured intellectual labor. The software industry is merely the first area to undergo this change, and once this model expands to sectors like law, finance, and design, the production relationships of knowledge work will be entirely rewritten.