Manish Kapur

Manish Kapur is vice president of product and solutions marketing at Sonar. His product writing focuses on code quality, security, and independent verification of AI-generated code: giving coding agents project-specific guidance, checking their changes as they work, and verifying automated repairs before engineers review them. In his co-authored introduction to Sonar Vortex, he explains why code generation and verification should use separate methods—the system that produces code can miss problems when judging its own output.

Kapur joined SonarSource as senior director of product marketing and ecosystem development after more than 13 years at Oracle. In his announcement of the move, he described an opportunity to contribute to the company’s Clean Code work through SonarLint, SonarQube, and SonarCloud, which help developers build secure, reliable, maintainable software. His subsequent writing at Sonar applies those concerns to workflows where agents generate and modify code.

Bringing code analysis into the agent’s workflow

Kapur announced the general availability of SonarQube MCP Server in October 2025. The server uses the Model Context Protocol to connect compatible AI assistants to SonarQube Cloud or Server. An assistant can retrieve code issues, request analysis of a file or snippet, and check whether a project passes its quality gate—the team’s configured conditions for acceptable code quality—within an editor or command-line workflow.

His explanation of Sonar’s Guide–Verify–Solve framework develops three connected priorities:

• Give agents relevant project context. Before changing code, an agent needs to understand the project’s architecture, coding standards, and security constraints. Kapur emphasizes selecting guidance relevant to the task rather than filling the agent’s context with every available rule. The aim is to help an agent produce changes that fit the surrounding system, rather than code that works only in isolation.

• Check changes while the agent is working. Waiting for a pull request or a failed continuous integration check delays feedback until the agent may have moved on. Bringing analysis into the generation loop lets the agent see and correct security, reliability, and maintainability issues while it still has the task’s context.

• Verify repairs before human review. Automated remediation can address both newly introduced issues and an existing technical-debt backlog. In the workflow Kapur describes, proposed fixes are scanned again to check that they resolve the original issue without introducing new ones. Engineers receive pull requests to review and retain control over what merges.

From open betas to Sonar Vortex

Kapur’s March 2026 article introduced open betas for Sonar Context Augmentation, SonarQube Agentic Analysis, and SonarQube Remediation Agent. In June, he and Prasenjit Sarkar co-authored the launch announcement for Sonar Vortex, which combined context augmentation and agentic analysis into one generally available product, alongside the general availability of the Remediation Agent.

The announcement gives a concrete account of how verification moves earlier in development. SonarQube stores project analysis context during a normal continuous integration run; Vortex restores that context when an agent needs to check a change, allowing analysis of individual files or groups of files without repeating the full scan. Architectural guidance helps the agent choose a change, analysis checks the result, and remediation produces verified fixes for engineers to assess.