Google’s Project Manager Ryan Salva Discusses the Evolution of AI in Coding and the Emergence of Agentic Programming

Google’s project manager for developer tools, Ryan Salva, shares insights into the transformative impact of AI tools on coding. With a background at GitHub and Microsoft, Salva now oversees tools such as Gemini CLI and Gemini Code Assist, guiding developers into the realm of agentic programming.

In a recent report released by his team, data was presented on how developers utilize AI tools and the advancements still to be made in this field. In an exclusive interview, Salva discusses the report’s findings and his personal experiences with AI coding tools.

The discussion was edited for brevity and clarity.

Annual developer trend surveys by Google place a significant focus on AI tools, particularly the adoption of agentic programming by developers. One of the intriguing discoveries from the research is the median date when developers began using AI tools – April 2024. This correlates with the release of Claude 3 and Gemini 2.5, marking a pivotal moment in the development of reasoning or thinking models and advancements in tool-calling capabilities.

For coding tasks, external information leveraging is crucial for problem-solving. This may involve grep commands, code compiling, unit testing, and integration testing. Tool-calling is a vital component that enables models to self-correct during the development process.

Salva uses AI coding tools personally, primarily for hobby projects, with command line-based tools like Gemini CLI, Claude Code, and Codex among his choices. He employs a variety of Integrated Development Environments (IDEs) such as Zed, VS code, Cursor, and Windsurf to observe the industry’s evolution.

On the professional front, product managers typically work within documents. Salva utilizes AI tools to help draft specifications and requirements documentation. He explains that he uses Gemini CLI to create more detailed requirement documents in Markdown format, which subsequently guide the coding process. The engineering team employs multiple layers of rules and Markdown docs, consumed by the model to adhere to the team’s workflow practices.

As Gemini CLI troubleshoots issues, it updates Salva’s requirements documentation detailing each step taken towards resolution. Each update creates a new commit and pull request within the repository, ensuring flexibility in reverting changes if needed. Approximately 70-80% of his work involves crafting requirements using natural language with Gemini CLI, allowing the tool to write most of the code, which he then reviews and edits using various IDEs primarily for reading purposes.

The question arises as to whether there is a future for raw computer code or if everything will transition into terminal windows. Salva reflects that, over three decades, Integrated Development Environments (IDEs) have been the hub of software development. However, he predicts that as AI tools advance, developers will spend more time focusing on requirements and less time in IDEs. This change may occur gradually over an extended period.

The evolution of software development might lead to questions about its impact on developers’ roles. Salva believes that the job profile of a developer will shift towards that of an architect, focusing on breaking down complex problems into manageable tasks and considering the bigger picture rather than intermediate machine code.