The AI Productivity Paradox in Engineering | Sonar Summit 2026

At the Sonar Summit 2026, Chris Grams, VP of corporate marketing for Sonar, hosted a fireside chat with Ganesh Datta, co-founder and CTO of Cortex, to explore the complex relationship between AI adoption and engineering outcomes. The discussion centered on a comprehensive research report examining whether organizations are realizing the promised benefits of AI coding assistance tools. Datta's background in fintech, where he witnessed the evolution from monolithic architectures to hundreds of microservices, provided crucial context for understanding modern engineering operations challenges and the need for better organizational maturity frameworks.

The Paradox: Faster Development, More Incidents

The report's central finding reveals a troubling disconnect: while organizations are shipping code approximately 20% faster with AI assistance, they are experiencing a roughly equivalent increase in incidents and quality regressions. This suggests that the long-anticipated productivity gains from AI coding tools are being offset by new quality challenges. Datta emphasized that this trend likely extends beyond the research period, hypothesizing that current organizations are probably experiencing even more pronounced versions of this pattern as AI adoption accelerates. The metric of incidents per pull request increased by 23.5%, indicating that not only are teams shipping more code, but the ratio of incidents to shipped code has worsened considerably.

Code Generation Was Never the Bottleneck

The discussion revealed a fundamental insight: writing code was never the primary constraint in software delivery. Datta argued that organizational processes, compliance requirements, testing protocols, and human decision-making have always been the real bottlenecks. AI tools have successfully accelerated the inner loop of code generation and reduced typing time, but they have not meaningfully improved the surrounding organizational infrastructure that governs how code flows through testing, review, compliance, and deployment. As a result, teams are now generating more code faster, but feeding it into organizational pipes that remain unchanged from pre-AI times, creating a capacity mismatch.

The Governance and Quality Assurance Gap

One significant finding from the research is that fewer organizations than expected have implemented governance policies around AI adoption. While AI coding assistance has achieved nearly universal adoption across large pockets of most organizations, the governance frameworks and quality assurance practices have lagged significantly behind. Datta emphasized the importance of alignment and measurement standards—organizations must collectively understand what "good" looks like and ensure teams are "all rowing in the same direction." The research highlights that psychological safety through testing and code coverage is critical not just for quality metrics, but for organizational confidence in shipping to production and recovering from issues when they inevitably arise.

Key Takeaways

• Speed-Quality Trade-off: AI coding tools are delivering a 20% increase in development speed but an equivalent increase in incidents and quality regressions, creating a paradoxical productivity gain
• Process, Not Code, is the Bottleneck: The shipping process and organizational governance structures are the real constraints in software delivery, not code generation itself
• Governance Lags Adoption: Most organizations lack formal governance policies for AI, despite widespread adoption of AI coding assistance tools across their teams
• Incidents Per PR Metrics Matter: The 23.5% increase in incidents per pull request suggests AI-generated code may be less understood and more complex, requiring stronger testing and code coverage practices
• Organizational Alignment is Critical: Success with AI requires teams to align on standards, measurement frameworks, and quality definitions rather than simply pursuing velocity gains