CodeRabbit scales agentic code reviews with OpenAI and ships 20x more

CodeRabbit, an AI-powered governance and control platform for code review, was built around a simple observation from CEO Harjot Gill: the first bottleneck every company hits after adopting coding agents is code reviews. The company says it is used by over 20,000 organizations and 150,000+ open-source projects, and its role is to help teams understand, review, secure, and decide what to ship from AI-generated code.

CodeRabbit‘s product has evolved with OpenAI models. It started with GPT-3.5 Turbo; Harjot says GPT-4 made the product effective at reviewing code, and the company later adopted OpenAI o1 and o3 early. Because code reviews run asynchronously in the cloud, CodeRabbit could afford slower reasoning models in exchange for more capable analysis. The team evaluates new models on its own benchmark of known issues; at the time of the interview, Harjot estimated that GPT-6 Astra caught 50% of the issues in that dataset, putting it at the top of models tested. The benchmark itself keeps evolving because agents introduce new mistake patterns and fixes reveal problems earlier reviews missed.

Better model price-performance and caching let CodeRabbit move capabilities from research into production. It started with one or two subagents for code navigation; lower-cost models such as GPT-6 Luna now support more subagents across the platform, while GPT-6 Astra handles planning and orchestration. Harjot says CodeRabbit is now using agents on use cases it could not previously afford, which expands the validation it can perform and the explanations it gives developers.

Using Codex, the company reports shipping roughly 20 times more software over the last six months without adding engineers, and Harjot estimates he pushed 2,500–3,000 pull requests in the 30 days before the interview. That required rethinking the release pipeline: running multiple agents exposed constraints in laptop memory, compute, testing, packaging tools, and dependencies. CodeRabbit streamlined releases, reconsidered where human approval was needed, and used its own product to automatically ship low-risk changes. It also built a Codex-powered Slack bot that triages production incidents, Slack messages, and support tickets and can return a pull request for review.

CodeRabbit uses Codex outside engineering as well, managing Google Ads through pull requests and using a content studio repository where Codex produces slide decks and customer-facing documents. Harjot describes the approach as running the company as code. On the security side, CodeRabbit used OpenAI‘s Cyber model through Daybreak for internal penetration testing and found multiple vulnerabilities it had not imagined. The capability is now part of CodeRabbit Secure, which uses Daybreak with Codex models to find and fix vulnerabilities; over two weeks, a promotional offer covered $1M in free deep scans, and the company says it found over 43,000 security vulnerabilities across 13,000 scans, thousands of which customers have already fixed.

Looking ahead, CodeRabbit is moving toward agentic software development, helping teams validate, prioritize, explain, secure, and verify agent-generated code at scale. As background agents begin responding to alerts and incidents without a human initiating every task, Harjot expects teams to need more support deciding which changes to take to completion, keeping developers in control of the code being produced.

CodeRabbit scales agentic software development and ships 20x more

View Original