Anirban Chatterjee is a product and solutions marketing leader at Sonar focused on independent code verification for AI-assisted software development. His central concern is that coding agents generate software faster than engineering teams can establish whether it is secure, maintainable, and ready for production.
Chatterjee began his career as a software developer at IBM, writing server-side code, before working across enterprise-software startups and moving into product leadership and marketing. At AI Engineer World’s Fair 2026, he represented Sonar as Director, Product and Solutions Marketing.
His work has followed AI coding from individual assistants to autonomous agents. In 2024, he outlined standards for evaluating AI-generated contributions. The following year, he described SonarQube quality gates for GitHub Copilot and introduced a program connecting SonarQube with development tools. By 2026, his focus included verification inside Claude Code workflows and automated remediation across software delivery.
- Verification debt: Agents create a growing liability when code production outpaces trustworthy review. Chatterjee argues that human reviewers can overtrust confident AI output, while production systems require stricter safeguards than disposable experiments. His analysis of Sonar’s developer survey identifies weak verification as a bottleneck in widespread AI-assisted development.
- State of Code: Chatterjee helped introduce Sonar’s research into software reliability and maintainability. His maintainability analysis highlights tests containing no assertions: apparent coverage rises even when meaningful verification does not.
- Guide, verify, solve: His agent-centric development cycle supplies agents with architectural constraints and relevant context, independently checks generated code, and returns detected problems for remediation. Verification operates both inside coding loops and through CI/CD quality gates; the model generating code cannot serve as its only judge.
Chatterjee sees stronger verification as a prerequisite for safely expanding agent autonomy. As he noted publicly, developers increasingly spend their time validating and correcting generated software, making dependable oversight essential to productive delegation.
