As Google’s project manager for developer tools, Ryan Salva oversees the development of AI-powered coding tools like Gemini CLI and Gemini Code Assist, helping shape how developers interact with artificial intelligence in their workflows. With a background at GitHub and Microsoft, Salva brings deep experience to the evolving landscape of agentic programming—where AI systems don’t just suggest code but actively plan, execute, and debug tasks. His team recently released third-party research highlighting how developers are actually using AI tools, revealing key insights into the shift toward more autonomous coding. One surprising finding was that the median date developers began using AI tools was April 2024—coinciding with the launch of advanced models like Claude 3 and Gemini 2.5. These models marked a turning point, particularly due to their improved ability to perform tool-calling, such as running commands, compiling code, or executing tests. This capability allows AI to self-correct and iterate, which is essential for solving real-world coding problems. Salva uses AI tools extensively in both his personal and professional work. For hobby projects, he relies on terminal-based tools like Gemini CLI, alongside others like Claude Code and Codex. He uses multiple IDEs—Zed, VS Code, Cursor, and Windsurf—not just to test different environments, but to understand how the industry is evolving. On the professional side, he uses AI to draft technical specifications and requirements documents, which are often vague or incomplete. He starts with a GitHub issue, then uses Gemini CLI to generate a detailed, outcome-driven specification in Markdown—around 100 lines of technical guidance. This document is enriched with team-specific rules on testing, dependency management, and coding standards, which the model uses as context. The AI then writes the code based on this specification. As the process unfolds, Salva has the AI update the requirements doc with progress notes—“I fixed this step, now moving to the next”—and each action creates a commit and pull request. This creates a transparent, traceable workflow that allows him to review, rewind, or undo changes easily. He estimates that 70% to 80% of his development work now happens through natural language in the terminal, with the IDE serving mainly as a code reader rather than a writing tool. Looking ahead, Salva questions whether raw code will remain central. While IDEs will still play a role, he believes developers will spend more time defining high-level goals and breaking down complex problems—shifting from writing code to architecting solutions. The job of a developer, he says, will evolve into that of a problem solver and planner, focusing on the big picture rather than syntax. He acknowledges concerns about the future of software development, but sees opportunity in this transformation. The role won’t disappear—it will change. Developers who can guide AI, set objectives, and evaluate results will remain essential. The future isn’t about replacing coders, but redefining what it means to be one.