# Cursor's AI Agents Enhance Code Verification and Merge Efficiency > With AI agents capable of autonomously merging pull requests, Cursor is reshaping the software development landscape. Their unique verification architecture highlights the critical role of evidence in ensuring coding quality and efficiency. **Source**: arize.com | **Published**: 2026-07-17 | **Type**: case_study ## Key Facts - Cursor's agents enable 30-40% of pull requests to merge without human review, enhancing efficiency. - Continuous risk scoring directs human review to high-impact changes, optimizing resource allocation. - The integration of video evidence from agents improves decision-making, revealing user-facing behavior. - A robust verification architecture allows for scalable automation, reducing defects and improving quality. - Cursor's skill pruning strategy enhances context relevance, ensuring agents operate with sharper focus. ## Summary \## Summary Cursor, a software development company, faced challenges in managing the increasing volume of pull requests generated by AI agents. To address this, they implemented a verification architecture that integrates automated risk scoring and human review processes. As a result, approximately 30 to 40 percent of pull requests now merge without human review, significantly streamlining their development workflow. \## Background Cursor operates in the software development industry and has developed a suite of AI agents to assist in code generation and review. Prior to the deployment of their verification architecture, the company struggled with the efficiency of their development process, particularly as the volume of pull requests surged due to the use of AI. \## Challenge The primary challenge was to manage the growing number of pull requests generated by AI agents while ensuring quality and minimizing the need for human intervention in routine changes. The company needed a system that could effectively assess the risk and quality of these automated changes. \## Solution Cursor developed a comprehensive verification architecture that spans the entire software development lifecycle. This system includes automated risk scoring, behavioral artifacts, and specialized review agents. The architecture allows agents to operate within developer-like environments, enabling them to implement changes and return evidence, such as videos demonstrating the functionality of new features. This evidence aids reviewers in making informed decisions about whether to merge changes. \## Results The implementation of this verification architecture has led to a significant increase in efficiency, with 30 to 40 percent of pull requests now merging without human review. This shift has allowed Cursor to streamline its development process, reducing the bottleneck typically associated with code reviews. \## Key Insights 1. Integrating automated verification processes can significantly enhance the efficiency of software development workflows. 2. Providing evidence of changes, such as video demonstrations, can improve decision-making and reduce the need for human oversight. 3. Continuous improvement of the system through feedback and corrections is essential for maintaining quality and trust in automated processes. \## Customer Testimonial “...that artifact marked one of the turning points in Cursor’s internal use of agents.” — John Gilhuly, Regional Director of Field Engineering at Cursor. ## Entities - **Companies**: Cursor, Arize, Spotify, Anthropic, OpenAI, Gemini - **Products**: Bugbot - **Technologies**: AI, CI, security review, risk scoring - **People**: John Gilhuly ## Key Concepts AI-written code verification, agent architecture, pull request automation, risk scoring, developer environment, Bugbot learning, skill management, evidence-based decision making ## Definitions - **CI**: Continuous Integration (CI) is a software development practice where code changes are automatically tested and merged into a shared repository. - **risk scoring**: Risk scoring is a method used to evaluate the potential impact of changes in code, determining the level of human review required. - **Bugbot**: Bugbot is Cursor's specialized agent for reviewing pull requests, learning from feedback to improve future evaluations. - **monorepo**: A monorepo is a single repository that houses multiple projects, allowing for shared code and resources across them. - **evidence contract**: An evidence contract outlines the necessary tests and artifacts required before an automated merge can occur. ## Use Cases - Security review on individual pull requests - Detection and repair of CI failures - Automatic fixes for issues found during code review - Test coverage maintenance - Continuous checks for documentation - Vulnerability confirmation and remediation workflows ## Frequently Asked Questions **What is the role of agents in Cursor's development process?** Agents in Cursor's development process automate tasks such as code changes and pull request reviews, allowing for faster and more efficient workflows. They gather evidence and provide risk assessments to determine the need for human review. **How does Cursor ensure the quality of AI-generated code?** Cursor employs a verification architecture that integrates CI, security reviews, and risk scoring to assess the quality of AI-generated code. This system allows for automated merges of low-risk changes while routing higher-risk changes to human reviewers. **What is Bugbot and how does it learn?** Bugbot is Cursor's pull request review agent that learns from human feedback and review comments. It incorporates corrections into its future evaluations, turning missed issues into regression cases for improved accuracy. **What are the benefits of using a monorepo?** Using a monorepo allows Cursor to manage shared code and skills across multiple projects, enhancing collaboration and reducing redundancy. It also simplifies the process of maintaining and updating code. **How does Cursor handle failures in the evaluation process?** When an evaluation fails, Cursor's system triggers a diagnosis workflow that investigates the issue using relevant logging context. This process helps identify and rectify problems, improving future evaluations. ## Links - [Read on Welcome.AI](https://welcome.ai/content/cursors-ai-agents-enhance-code-verification-and-merge-efficiency) - [Original source](https://arize.com/blog/inside-cursors-agent-factory-how-it-verifies-ai-written-code/) - [Arize AI](https://welcome.ai/company/arize-ai): Featured company --- Source: Welcome.AI | https://welcome.ai/content/cursors-ai-agents-enhance-code-verification-and-merge-efficiency