CodeRabbit’s Governance Platform
CodeRabbit has developed a governance platform that reviews, validates, and secures code generated by AI agents. The company attributes significant improvements in software delivery to the integration of OpenAI models and Codex. Currently, CodeRabbit is utilized by over 20,000 organizations and more than 150,000 open-source projects.
Scaling Code Review with AI Models
According to Harjot Gill, co-founder and CEO of CodeRabbit, the primary challenge teams face when adopting coding agents is the code review process. CodeRabbit’s platform facilitates asynchronous, cloud-based reviews, allowing the use of slower yet more capable models for in-depth code reasoning. The platform initially utilized GPT-3.5 Turbo, became more effective with the introduction of GPT-4, and later adopted OpenAI o1 and o3 models. CodeRabbit continuously evaluates new models against an internal benchmark of known issues, with Harjot estimating that GPT-6 Astra identified approximately 50% of the issues in their dataset.
Improving Delivery and Operations
By incorporating Codex into its workflow, CodeRabbit reports shipping around 20 times more software over a six-month period without increasing its engineering team size. Harjot estimated that he personally submitted between 2,500 and 3,000 pull requests in the month leading up to the report. The company has also restructured its release processes, reassessed the necessity of human approvals, and automated low-risk changes.
Utilizing Codex Beyond Engineering
Beyond engineering tasks, CodeRabbit employs Codex to manage marketing and operational activities through code. This includes managing Google Ads via pull requests and utilizing a content studio to create slide decks and documents. The company has implemented agents in sandbox environments and has leveraged OpenAI’s Cyber model through Daybreak for internal penetration testing.
Enhancing Security with CodeRabbit Secure
CodeRabbit’s security solution, CodeRabbit Secure, integrates Daybreak with Codex-based agents to identify and rectify vulnerabilities. Recently, during a promotional campaign offering $1 million in free deep scans, the company discovered over 43,000 security vulnerabilities across approximately 13,000 scans. Thousands of these vulnerabilities have already been addressed by customers, significantly enhancing their code security.
Original source: OpenAI News