How Google’s dev tools manager makes AI coding work

Ryan Salva, who leads developer tools at Google, is at the forefront of a revolution in coding driven by AI technologies. Previously with GitHub and Microsoft, Salva is now overseeing tools like Gemini CLI and Gemini Code Assist, pushing developers toward a new era of agentic programming. In a recent report, Salva's team unveiled third-party research shedding light on how developers are integrating AI tools into their workflows, and the substantial progress that still lies ahead. I had an opportunity to discuss these findings with Salva, who shared his insights and personal experiences with AI in coding. This year’s edition of Google’s developer trends survey placed a significant focus on AI tools, particularly examining how willing developers are to embrace agentic programming. One compelling revelation from the research was the median date when developers began utilizing AI tools: April 2024. This coincides with the release of Claude 3 and Gemini 2.5, marking a pivotal moment in the evolution of reasoning models and tool-calling capabilities. Salva emphasized that effective coding often relies on harnessing external information to solve problems, whether that involves searching for data or compiling code. This tool-calling aspect is essential, allowing models to self-correct during the coding process. On a personal level, Salva primarily engages in hobby projects, frequently using command-line tools like Gemini CLI. He mentioned that his coding environment is diverse, incorporating various integrated development environments (IDEs) such as Zed, VS Code, and Cursor. His professional responsibilities also involve leveraging AI to streamline the creation of specification and requirements documents. When discussing the role of Gemini CLI in his workflow, Salva explained that a typical development task starts with a GitHub issue, which often lacks specificity. He utilizes Gemini CLI to enhance the requirement documents in Markdown format, generating detailed specifications that guide his coding efforts. This process not only helps in writing code but also in maintaining thorough documentation of changes through commits and pull requests. Regarding the future of coding, Salva speculated that the traditional IDE might gradually lose its central role. He anticipates a shift where developers will focus more on requirements and high-level problem-solving rather than directly interacting with raw code. This evolution could redefine the role of developers, making it more akin to that of architects, who break down complex problems into manageable tasks, emphasizing the overall vision rather than the intricate details of programming languages.