Google is officially shifting the long standing software development paradigm. During his keynote at Cloud Next 2026, Sundar Pichai revealed that a staggering 75% of all new code at Alphabet is now generated by artificial intelligence before being passed to engineers for approval. This is a drastic leap – just last fall, that figure stood at 50%, and in October 2024, it was a mere 25%.
From an operational perspective, this marks a transition from “assisted coding” toward “agentic workflows.” AI is no longer just an overlay on an engineer’s work, but the default code creator. Humans have been shifted to the roles of reviewers, orchestrators, and quality control supervisors.
The effects of this approach are already measurable. As Google reported, one internal software migration task, conducted using more autonomous AI workflows, was completed six times faster than under the standard model.
Massive investments and hard data
This technological pivot is built on a powerful financial foundation, which the board reported during the earnings call in February 2026. Investments in AI are clearly translating into the corporation’s overall performance:
- Alphabet revenues (Q4 2025): $113.8 billion (an 18% year-over-year increase).
- Google Cloud: Revenues reached $17.6 billion, and operating income in this sector doubled, hitting $5.3 billion.
- Capital Expenditures (Capex): In 2025, the company spent $91.4 billion (with $27.9 billion in Q4 alone). Forecasts for 2026 project astronomical spending in the range of $175–185 billion, mostly earmarked for AI technical infrastructure.
Internal divisions and the rise of Anthropic
Despite the impressive official statistics, internal reports paint a slightly more complex picture. According to Bloomberg, Google is trying at all costs to unify its internal developer initiatives amid mounting fears that competitors – Anthropic in particular – are gaining the upper hand in the market for AI developer tools.
The situation is further complicated by reports from Business Insider. According to them, certain Google DeepMind teams have been granted permission to use Claude Code (a rival product from Anthropic), even though the vast majority of the company is required to rely on internal tools based on the Gemini model. This is causing friction within the corporation regarding tool quality and access.