AI: Writes 75% of Google's new code, yet 46% of developers distrust it, and the other figure is being ignored.
google
| Source: Dev.to | Original article
AI now generates three quarters of Google's new code, yet less than half of developers trust its output.
Google’s engineering chief announced that AI now writes three‑quarters of the company’s new code, a jump from roughly a quarter in October 2024. The figure comes from internal metrics that also show human engineers reviewing every AI‑generated contribution. While the rollout is billed as a productivity boost – Google cites a 10 percent uplift across its engineering teams – a parallel internal survey reveals that 46 percent of developers distrust the output they receive from the tools.
The shift matters because it signals a tipping point for the “agentic coding” era that has been discussed in industry circles for years. If AI can produce the bulk of a firm’s software, the role of human programmers moves from writing to supervising, testing and fixing. Yet the same data underline a security gap: AI‑written code passes Google’s automated security checks only 56 percent of the time, according to Axis Intelligence Research. The disparity between volume and reliability raises questions about how quickly firms can safely scale AI assistance without exposing systems to new vulnerabilities.
The development builds on trends we highlighted earlier this month in “AI Can Write More of the Code. We Still Have to Engineer the System,” where we noted that AI‑generated code now accounts for 41 percent of global software output, with Google leading the pack. Watching how Google tackles the trust deficit – whether through stricter review pipelines, new tooling, or changes to developer training – will be key. Equally important will be the response from rivals such as Meta, which has announced similar ambitions, and the broader industry’s move toward redefining engineering productivity metrics in an AI‑dominant landscape.
Sources
Back to AIPULSEN