Google’s project manager for developer tools, Ryan Salva, provides unprecedented insights into how AI coding tools are fundamentally changing software development. With experience from GitHub and Microsoft, Salva now leads groundbreaking tools like Gemini CLI and Gemini Code Assist, pushing developers toward agentic programming. Recently released research reveals exactly how developers utilize AI coding tools and what progress remains.
The Dawn of Intelligent AI Coding Systems
Google’s latest developer trends survey focuses specifically on AI coding adoption. Surprisingly, research shows the median start date for developers using AI coding tools was April 2024. This timing coincides with major advancements including Claude 3 and Gemini 2.5 releases. Consequently, we’ve entered the era of reasoning models with significantly improved tool-calling capabilities.
For coding tasks, models must leverage external information to problem-solve effectively. Therefore, they need to:
- Execute grep commands and compile code
- Run unit tests and integration tests
- Self-correct throughout the development process
Personal AI Coding Workflow Implementation
Salva personally uses multiple AI coding tools for hobby projects and professional work. He primarily utilizes command-line tools like Gemini CLI while maintaining flexibility across various IDEs. Importantly, his heterogeneous approach includes Zed, VS Code, Cursor, and Windsurf. This strategy allows him to monitor industry evolution comprehensively.
On the professional side, product managers typically work extensively with documents. Salva uses AI coding tools to:
- Create detailed specification documents
- Develop robust requirement documentation in Markdown
- Generate technical, outcome-driven specifications
Building AI Coding Tools with AI Assistance
Salva’s team employs an innovative approach where they use Gemini CLI to enhance Gemini CLI itself. Development tasks typically begin with underspecified GitHub issues. The team then uses AI coding tools to transform these into comprehensive requirement documents. Subsequently, the system generates code based on specifications and team preferences.
The engineering team maintains multiple layers of rules and Markdown documentation that guide the AI coding process. These documents outline:
- Testing methodologies and dependency management
- Team-specific working patterns and standards
- Quality assurance protocols and best practices
The Future Landscape of AI Coding Development
For three decades, IDEs dominated software development workflows. However, AI coding advancements suggest a significant shift toward requirement-focused work. Consequently, time spent in traditional IDEs will likely decrease gradually over an extended horizon.
This transformation raises important questions about software development careers. Nevertheless, Salva believes developers will evolve into architectural roles. They’ll focus on breaking complex problems into solvable tasks rather than writing intermediate language code. Ultimately, AI coding will elevate developers to higher-level problem-solving positions.
Frequently Asked Questions
What is agentic programming in AI coding?
Agentic programming refers to AI systems that can autonomously execute complex coding tasks. These systems leverage tool-calling capabilities to self-correct and complete multi-step development processes.
How does Gemini CLI differ from traditional coding tools?
Gemini CLI operates primarily through natural language commands in the terminal. It generates code based on specifications and team documentation, reducing direct IDE usage for code writing.
When did AI coding tools become widely adopted?
Research indicates April 2024 marked the median adoption date. This period coincided with major model releases that significantly improved reasoning and tool-calling capabilities.
Will AI coding eliminate developer jobs?
Rather than eliminating positions, AI coding will transform developer roles toward architectural functions. Professionals will focus on problem decomposition and high-level design rather than manual coding.
What IDEs work best with AI coding tools?
Modern AI coding tools maintain compatibility with multiple IDEs including VS Code, Zed, Cursor, and Windsurf. The choice depends on specific project requirements and team preferences.
How do AI coding tools handle testing and quality assurance?
These tools integrate testing protocols through documented guidelines. They can execute unit tests, integration tests, and update requirement documentation throughout the development lifecycle.
