Google CEO Sundar Pichai Says AI Is Transforming Software Development

Google CEO Sundar Pichai has highlighted the growing influence of artificial intelligence on software development, explaining that AI tools are increasingly changing how engineers build and maintain the company’s products. In April 2026, Pichai revealed that approximately 75% of Google’s newly written code was being generated by AI and subsequently reviewed and approved by human engineers, compared with 50% in the previous autumn. More recently, he has suggested that measuring AI adoption solely by the percentage of code it generates is becoming less useful because the technology is now influencing almost every stage of software development. The shift reflects Google’s broader ambition to make AI an essential part of its engineering process rather than simply a tool for completing small programming tasks.

A major development in this strategy is Google’s increasing use of AI agents capable of handling complex engineering assignments. Unlike traditional coding assistants, which primarily suggest lines of code or complete short functions, these agents can work through larger tasks, make changes across multiple files and assist with complicated software migrations. Pichai previously reported that a particularly complex code migration completed by engineers working alongside AI agents was finished six times faster than a comparable task could have been completed by engineers working alone a year earlier. Google sees these capabilities as an opportunity to help its workforce tackle more ambitious projects while reducing the time spent on repetitive programming work.

Google has also begun evaluating AI-assisted development through practical engineering outcomes rather than simply counting how much code AI produces. According to reporting published on October 21, Pichai highlighted improvements associated with the company’s internal Argon coding agent. During the first four weeks following its rollout, Google reportedly recorded a 30% increase in code submissions, alongside a reduction of approximately one-third in code rollbacks. Code submissions measure proposed changes to software, while rollbacks occur when previously submitted changes must be reversed. These figures suggest that AI-assisted development could increase the volume of engineering work while reducing some forms of rework. However, the results were reported by Google, and the figures alone do not establish how much of the improvement was caused exclusively by AI.

The transformation also raises questions about the future responsibilities of software engineers. As AI becomes more capable of generating and modifying code, engineers may spend less time writing every line manually and more time designing systems, reviewing proposed changes, testing software and identifying security vulnerabilities. Human oversight remains important because AI-generated code can contain bugs, introduce security weaknesses or misunderstand the intended requirements of a project. Google’s approach retains engineers in the approval process, recognising that producing more code does not automatically mean producing better software. The challenge for technology companies will be to ensure that increased development speed is matched by reliable testing, effective security reviews and clear accountability for changes made with AI assistance.

Pichai’s comments come as competition intensifies among major technology companies seeking to integrate AI throughout their businesses. Google is developing its Gemini models and coding agents, while Microsoft, OpenAI and Anthropic are also investing heavily in AI-powered software development. These companies increasingly view AI agents as systems capable of completing multi-step tasks, rather than tools that simply answer questions or generate text. For Google, the strategy also provides an opportunity to use its own technology internally before offering similar capabilities to enterprise customers. If the reported productivity gains can be sustained across larger projects, AI-assisted engineering could influence how software is developed across the technology industry. Nevertheless, the long-term impact will depend on whether these systems consistently deliver secure, maintainable and dependable results.