CodeRabbit, an AI code review vendor, says it shipped roughly 20 times more software over six months without adding engineers using OpenAI’s Codex, CEO Harjot Gill has revealed.
Gill also estimated he personally pushed 2,500 to 3,000 pull requests in the 30 days before his interview. A pull request proposes a change to a codebase for review and possible inclusion.
How CodeRabbit changed software delivery
According to the firm, running multiple coding agents exposed problems throughout CodeRabbit’s development process. The company encountered limits in laptop memory and computing capacity, along with bottlenecks in testing, software-building tools and dependencies. Reviewing the growing volume of code was another challenge.
“The first bottleneck that every company hits as soon as you adopt coding agents is code reviews,” Gill said.
CodeRabbit responded by streamlining its release process and reconsidering which changes needed human approval. It now uses its own review software to automatically ship changes classified as low risk. A separate Codex-powered Slack bot triages production incidents, Slack messages and support tickets, then prepares pull requests for employees to review and merge.
The company has extended pull-request workflows beyond engineering. Its marketing team uses them to deploy Google Ads, while a separate repository allows Codex to generate slide decks and customer documents. Gill argued that converting marketing tasks into code-based work could make them 10 times faster, although no measured results were provided.
Lower model costs and better caching have separately allowed CodeRabbit to deploy more AI agents in its commercial code-review product. Gill said CodeRabbit initially used one or two subagents for code navigation but can now afford to deploy more throughout its platform.
More code, different results
Research from Google Cloud’s DevOps Research and Assessment (DORA) program offers a separate view of AI-assisted development. Its 2025 report drew on responses from nearly 5,000 technology professionals and associated higher AI adoption with both greater software delivery throughput and greater instability. In a March 2026 analysis, DORA researchers wrote that “the time saved in creation is frequently re-allocated to auditing and verification.”
A July 2026 study published by the Association for Computing Machinery examined 40,214 pull requests across 2,807 GitHub repositories. Agent-authored requests had lower overall merge rates than human-authored ones, although those that were accepted were merged faster.