In modern microservices and containerized environments, service dependencies and security policies play a critical role in ensuring code functions cohesively across distributed systems. As software development grows more complex, code reviews increasingly require a “contextual” approach, where reviews consider not just the current pull request but also historical data, institutional knowledge, personal coding styles, and system-wide dependencies.
CodeRabbit-a new startup with the mission to bring AI to code reviews-is addressing this challenge by integrating AI into the existing software development workflows enhanced with institutional knowledge, such as personalized code styles, project management tools and ticketing systems, with both static datasets like linters and the dynamic reasoning capabilities of large language models (LLMs).
In this Q&A with VMblog, Gur Singh shares his thoughts on how CodeRabbit is using advancements in AI to make code reviews more accurate, automated, and context-aware to free up the manual time spent by developers in code reviews and elevate code quality.
VMblog: What’s broken with the code review process today?
Gur Singh: If you look at the entire CI/CD pipeline, it’s mostly automated today. But the code review process is one of its most manual, expensive areas. Manual code reviews are a very slow process. There was a recent study by GitLab that said on average a merge request took between 2-4 days to be merged into the main codebase. This doesn’t just slow down the developers, it slows down the project deliverables. It slows down the bug fixes and you cannot get code shipped out fast enough to meet business needs. Along with that, there is a lot of context switching required. Once the developer creates a pull request they switch into a new task. Then they get review feedback from their peers, and have to move backwards to fix issues discovered during reviews. We’re talking about a lot of time and pain for developers to go through manual code reviews.
VMblog: Why are traditional static tools like linters insufficient for modern code reviews?
Singh: Linters and static analyzers operate with fixed rule sets, flagging violations against predefined criteria. However, they cannot dynamically assess how changes in one file might impact other files or understand historical dependencies to better understand why certain code changes are being introduced. As coding environments become more dynamic, these tools lack the holistic reasoning required to address modern complexities.
VMblog: How does CodeRabbit handle institutional knowledge during code reviews, and enhance the context behind code changes before integrating with the LLMs?
Singh: CodeRabbit enriches code review context by pulling from various institutional knowledge sources, such as past JIRA tickets, previous pull requests, and comments from other developers. For example, if certain unit tests were required in a past ticket but changes in the current pull request result in needing to add new test cases that were missed, CodeRabbit highlights this gap, ensuring that code reviews account for both current and historical requirements. It gathers from workflow systems, tries to understand the intent of changes in code, looks for linked issues especially in Jira or Linear issue. It then runs the reviews using state-of-the-art LLM models with several recursive requests to reduce changes of false positives, parses responses, suppresses noise, and then presents an entire feedback review within the comments in the pull request. It does this automated, context aware code review whenever there is a new pull request.
VMblog: What is the relationship between CodeRabbit’s workflow and the human operator?
Singh: We’re not removing humans out of the loop. CodeRabbit works within the developers existing Git platform, finds bugs and defects or opportunities to improve code quality, and allows users to make changes with just 1-click. What we’re doing is reducing the manual parts of workflow, where studies show as much as 15-20 percent of engineering time goes into manual review, either reviewing or incorporating feedback found during reviews. CodeRabbit is giving much of that time back to developers so they can use it to build additional features of value to their company. CodeRabbit also enhances code quality by providing refactor suggestions, automatically adding unit-test cases and finishing touches to make the code more robust and maintainable. CodeRabbit is helping companies move faster with code reviews and speed up their time to market by reducing and fixing the hard errors faster, with a consistent method that goes beyond static analyzers and linters that don’t understand the context behind complex code changes.
##
Guritfaq (Gur) Singh is the co-founder of CodeRabbit. He previously worked at Alegeus in Development. Gur attended Guru Nanak Dev University.
