Introduction
Artificial intelligence is revolutionizing software development in ways that seemed like science fiction just a short time ago. Ryan Salva, Google's senior director of product management for developer tools, offers a unique perspective on this epochal transformation. With a background including experiences at GitHub and Microsoft, Salva is now responsible for innovative tools like Gemini CLI and Gemini Code Assist, which are guiding developers toward a new era of agentic programming.
Google's Research: When Developers Embraced AI
Recent research conducted by Google revealed a surprising fact: the median date when developers started using AI tools is April 2024. This timing coincides perfectly with the release of Claude 3 and Gemini 2.5, models that marked the dawn of the "reasoning models" era.
According to Salva, this moment represents a crucial turning point: "For coding tasks, you really need to be able to leverage external information in order to problem solve. It may need to grep, it may need to compile the code, and if the code compiles it may want to run that unit test, and that integration test."
Salva's Personal Approach to AI Tools
The Google manager personally uses a variety of tools for his hobby and professional projects. His routine mainly includes command line-based tools like Gemini CLI, integrated with different IDEs such as Zed, VS Code, Cursor, and Windsurf.
"70% to 80% of my work is me working in the terminal with natural language, trying to use Gemini CLI to craft the requirements, and then allowing Gemini CLI to write most of the code for me."
Ryan Salva, Senior Director Google
The Transformed Development Process
Salva describes a completely new workflow: starting with an often under-specified GitHub issue, using Gemini CLI to create robust Markdown documentation of about 100 lines, and then letting AI write code based on technical specifications and team documents that define testing methodologies and dependency management.
Each troubleshooting step is automatically documented, with updates to requirement documents and creation of separate commits and pull requests, always ensuring the ability to rewind or undo changes.
The Future of Coding: Less IDE, More Architecture
Salva predicts a fundamental change in the developer's role: "Your job as a developer is going to look a lot more like an architect. It is going to be about taking big, complex problems and breaking them down into smaller, solvable tasks."
This evolution doesn't eliminate the programmer figure but transforms it. Future developers will focus more on requirements and the bigger picture, spending less time directly writing code in the IDE.
Conclusion
The transformation of coding through AI is no longer a future possibility but a present reality that is redefining the entire software industry. With increasingly sophisticated tools and an increasingly agentic approach, developers are evolving from code writers to solution architects, maintaining strategic control while delegating tactical implementation to artificial intelligence.
FAQ
When did developers start using AI tools for coding?
Google's research indicates April 2024 as the median date, coinciding with the release of advanced models like Claude 3 and Gemini 2.5.
What AI tools does Google's Ryan Salva personally use?
He mainly uses Gemini CLI, Claude Code, and Codex, integrated with IDEs like Zed, VS Code, Cursor, and Windsurf.
Will AI coding completely replace programmers?
No, it will transform their role from code writers to solution architects, focusing on complex problems and strategic vision.
How does the AI development workflow work according to Google?
It starts with under-specified requirements, uses AI to create robust documentation, then lets AI write code based on technical specifications.
Why is tool-calling important for AI coding?
It allows models to leverage external information, compile code, run tests, and self-correct during the development process.
How much time does Ryan Salva spend coding with AI tools?
70-80% of his work happens in the terminal with natural language, using IDEs mainly to read the generated code.