AnswersPublished15 sources
Why Google’s Sundar Pichai Says AI-Written Code Is No Longer a Useful Metric
Pichai says a share of AI authored code misses AI’s broader influence on engineering. Google’s earlier figures tracked AI generated code, including a reported 75% of new code generated by AI and approved by engineers in April 2026.
Published byEdited with GPT-6 LunaImages generated with GPT Image 2
Research answer
Key takeaways
- Pichai says a share of AI authored code misses AI’s broader influence on engineering.
- Google’s earlier figures tracked AI generated code, including a reported 75% of new code generated by AI and approved by engineers in April 2026.
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: Why did Google CEO Sundar Pichai tell Alex Kantrowitz in the October 20, 2026 Big Technology Podcast interview that the percentage of Google. Article summary: Pichai’s point was that “AI-written code” no longer captures AI’s contribution when it influences nearly the entire engineering process; Google is shifting its emphasis toward engineering impact rather than authorship pe. Topic tags: general, general web, user generated, news. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts w
Sundar Pichai’s answer was that a single percentage no longer captures how AI contributes to Google’s engineering work. In the October 20, 2026, Big Technology Podcast interview, he said the figure had become “meaningless” because AI now deeply influences every part of the work engineers touch. He pointed instead to code submissions and rollbacks after Google introduced Gemini 4 Argon. 5
Why the percentage became less useful
A share of AI-written code tries to answer a narrow question: how much code did AI generate? It does not show whether a team completed more useful work, spent less time reviewing changes, or shipped more reliable software.
That distinction matters when AI is involved in more than generating lines of code. If it also assists across a broader engineering process, separating work neatly into “AI-written” and “human-written” becomes less informative. Pichai’s point was about the limits of the measure—not proof that AI has improved every engineering outcome. 5
Google’s reported code figures show adoption, not impact
Earlier public figures offered a straightforward way to describe AI’s growing role. Google’s reported numbers rose from about 25% of code in late 2024 to 50% in fall 2025 and 75% in April 2026. The later figure referred to new code generated by AI and approved by engineers. 5
24
25
Those figures are not necessarily identical measures: the 75% figure specifically concerns new code that engineers approve. And even a rising share of AI-generated code establishes adoption, not by itself a rise in completed work, a reduction in engineering effort, or an improvement in software quality.
What Google says Argon changed
In the interview, Pichai said that during Argon’s first four weeks Google saw a 30% increase in “CLs”—code-change submissions—and roughly one-third fewer rollbacks. 5
Separately, Google said teams using Argon through Antigravity increased agentic code submissions by more than 35% and reduced code rollbacks by nearly a third. 45
51 The interview’s 30% figure and Google’s more-than-35% figure use different labels. The available reporting does not explain how their definitions or measurement populations compare, so they should not be treated as interchangeable.
Google has also described using Argon on complex codebase work. One example involved separating core webpage handling from a legacy Chromium folder of about six million lines. That illustrates the scale of a project the company says Argon assisted with; it does not mean the model generated six million lines of new code. 51
The caveat: submissions are not the whole productivity story
Code submissions and rollbacks are closer to engineering outcomes than an authorship percentage: one tracks submitted changes, while the other reflects changes that had to be undone. But they still do not provide a complete accounting of productivity. The figures are company-reported, and the available sources do not show a full comparison of completed work, review time, total engineering effort, and reliability.
So the shift is best understood as a change in what Google highlights. The earlier percentage described how widely engineers used AI to generate code; the newer measures aim to show what happened to submissions and rollbacks. Neither set of figures, on its own, establishes the total effect on software delivery. 5
45
51
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "Why Google’s Sundar Pichai Says AI-Written Code Is No Longer a Useful Metric"?
Pichai says a share of AI authored code misses AI’s broader influence on engineering.
What are the key points to validate first?
Pichai says a share of AI authored code misses AI’s broader influence on engineering. Google’s earlier figures tracked AI generated code, including a reported 75% of new code generated by AI and approved by engineers in April 2026.
Sources
bigtechnology.comSundar Pichai: Google’s Frontier Imperative, Compute Scarcity ...
bitsminds.comGoogle CEO: 75% of All New Code at Google Is Now AI-Written
metaintro.comGoogle's 75% AI Code Milestone: What Software Engineers Need to ...
reuters.comGoogle launches flagship AI model Gemini 4 Argon for developers, businesses