How CodeRabbit Scaled Agentic Software Delivery Using AI
Discover how CodeRabbit utilizes advanced AI models to review code, scale software delivery by 20x, and secure applications against vulnerabilities.
As organizations increasingly adopt autonomous coding tools, development teams face a new kind of hurdle: managing the surge of AI-generated work. Harjot Gill, co-founder and CEO of CodeRabbit, notes that the primary bottleneck shifts immediately to code reviews. To address this, CodeRabbit developed a governance and control platform that helps over 20,000 organizations and 150,000 open-source projects understand, review, and secure code produced by AI agents.
CodeRabbit's product evolution mirrors the rapid progress of large language models. Starting with GPT-3.5 Turbo, the company advanced to GPT-4, and later adopted reasoning models like OpenAI o1 and o3. Because code reviews operate asynchronously in the cloud, the platform can leverage models that require more time to reason through complex problems. During internal evaluations against a benchmark of known issues, newer models like GPT-6 Astra achieved high detection rates. Improved pricing and caching from providers have also allowed CodeRabbit to expand its use of subagents, employing models like GPT-6 Luna for code navigation and Astra for planning.
Shipping 20 times more software
By integrating Codex into its workflow, CodeRabbit managed to ship roughly 20 times more software over a six-month period without expanding its engineering headcount. Harjot Gill estimated that he personally pushed between 2,500 and 3,000 pull requests in the 30 days preceding the interview. Achieving this velocity required a complete overhaul of the software delivery lifecycle. The team streamlined release processes, automated low-risk changes using CodeRabbit itself, and addressed hardware and compilation bottlenecks.
Additionally, CodeRabbit built a Codex-powered Slack bot that handles production incidents, support tickets, and Slack messages by generating pull requests for human review. Beyond engineering, the team applies code-based workflows to marketing and operations, managing Google Ads and generating customer-facing documents through code repositories.
Securing code and managing vulnerabilities
Security remains a central focus as agentic workflows expand. CodeRabbit utilizes OpenAI's Cyber model via Daybreak for internal penetration testing, uncovering complex vulnerabilities that the team previously could not imagine. These capabilities feed directly into CodeRabbit Secure, a product designed to find and fix bugs. During a recent promotional offer featuring $1M in free deep scans, the platform uncovered over 43,000 security vulnerabilities across 13,000 deep scans using Codex models, leading to thousands of fixes by users.
What it means for developers
As background agents begin addressing operational alerts and writing code autonomously, developers need robust systems to maintain visibility and control. Platforms like CodeRabbit highlight how reasoning models can handle asynchronous tasks, from deep code reviews to automated vulnerability patching. For engineers looking to experiment with these capabilities, developers can try top AI models cheaply through one API at https://apixoai.online. Ultimately, the future of agentic development relies on keeping engineers in charge through transparent validation, detailed explanations, and automated security checks that scale alongside growing code volumes.
Source: CodeRabbit scales agentic software development and ships 20x more — OpenAI News. Written by the Apixo team from that report.
One key for Claude, GPT, GLM, DeepSeek and more. Pay per token with crypto.
Get your API key
