CodeRabbit scales agentic software development and ships 20x more

CodeRabbit, led by co-founder and CEO Harjot Gill, addresses the primary challenge faced by engineering teams adopting AI coding agents: code review bottlenecks. The company’s AI-powered platform validates and secures AI-generated code, supporting over 20,000 organizations and 150,000 open-source projects. By leveraging OpenAI models, CodeRabbit enables teams to understand, review, and ship AI-produced code with confidence, reducing manual oversight requirements.

Harjot Gill highlights that improved price-performance from OpenAI models allows CodeRabbit to deploy more AI agents for reviews, passing efficiency gains to customers. The company’s adoption of models like GPT-4, OpenAI o1, and OpenAI o3 enhanced its ability to analyze code asynchronously in the cloud. CodeRabbit’s internal benchmark shows GPT-6 Astra catching 50% of known issues, outperforming other models, while the benchmark evolves alongside emerging agent-generated bugs.

CodeRabbit expanded its use of OpenAI models to include Codex, enabling the company to ship approximately 20 times more software over six months without increasing its engineering team. Harjot Gill estimates he personally contributed 2,500 to 3,000 pull requests in the 30 days prior to the interview. This shift required rethinking software delivery processes, automating low-risk changes, and redistributing human approvals to maintain control over high-risk modifications.

Beyond engineering, CodeRabbit applies Codex to operational and marketing tasks, such as managing Google Ads via pull requests and generating slide decks through code. The company also uses OpenAI’s Cyber model through Daybreak for internal penetration testing, identifying previously unimaginable vulnerabilities. Its latest product, CodeRabbit Secure, leverages these capabilities to scan and remediate security flaws in customer code, having recently completed over 13,000 deep scans and identified 43,000 vulnerabilities.