Harness has redesigned its Git repository infrastructure to handle the explosive growth of AI-generated pull requests that traditional platforms struggle to support. This new system aims to maintain deployment velocity, reduce testing overload, and improve review focus for development teams worldwide.
- New repository designed for nonstop AI-driven pull request traffic
- Improved reviewer efficiency through context-aware prioritization
- Kubernetes-based infrastructure supports higher reliability and scale
Infrastructure signal
Harness rebuilt its Git repository infrastructure from the ground up to manage massive increases in pull request traffic driven by AI code generation tools. The new platform is Kubernetes-based and designed to operate 24/7 to support AI agents running continuously, unlike traditional repositories geared toward human work schedules. This change reflects a shift from repositories optimized for small teams operating during limited hours to systems built for nonstop, high-volume workloads.
By moving away from legacy single-tenant Git hosting models and embracing cloud native scalability, Harness's infrastructure aims to prevent outages and performance degradation seen in platforms like GitHub during high concurrency stress events. This rebuilt repository ensures more reliable handling of AI-driven source control operations while optimizing cloud consumption by avoiding wasted compute on irrelevant or low-priority pull requests.
Developer impact
The surge in pull requests generated by AI coding agents has overwhelmed testing and review teams, sometimes multiplying code volumes by 10x or more, pushing manual processes beyond sustainable limits. Harness’s new repository includes AI-enhanced code review capabilities that intelligently prioritize changes requiring human attention, using a software delivery knowledge graph to add context. This reduces noise from large dependency updates or scaffolding code, allowing reviewers to focus on critical functional changes.
Teams benefit from a more targeted review process where reviewers are matched based on prior interaction with affected code, not simply by who initiated the change. The improved workflow saves significant manual review time—Harness estimates over 10,000 hours monthly across customers—and enables higher deployment throughput without compromising quality or raising risk acceptance.
What teams should watch
Development and DevOps teams should monitor how increased AI-driven pull requests affect their continuous integration and deployment pipelines, focusing on bottlenecks in testing capacity and review throughput. Adopting infrastructure and tooling aligned with AI-first repository designs can help mitigate these challenges by scaling operations and improving prioritization of review tasks.
Observability should shift toward monitoring AI agent activity patterns and pull request characteristics, enabling dynamic adjustments in resource allocation and human review involvement. Teams must also evaluate database impact, such as locking or indexing issues discovered through incident reviews integrated into delivery pipelines. The emerging model of combining AI and knowledge graphs to contextualize code changes signals a platform-level transformation worth close attention.