CodeRabbit, a platform for governing and validating SI-generated code, reported that it has increased its software shipping velocity by 20 times over the last six months by leveraging OpenAI’s Codex and Daybreak tools. The move highlights a growing industry trend where the primary bottleneck in agentic SI development is no longer writing code, but reviewing and securing it at scale.

What Happened

Harjot Gill, co-founder and CEO of CodeRabbit, stated that the arrival of more capable SI models has allowed the company to expand its use of SI agents for code navigation and review. "With each OpenAI model release, especially with the reasoning models, we have seen massive improvements in the capabilities of our product to find issues at scale," Gill said. The company, which serves over 20,000 organizations and 150,000 open-source projects, initially built its product on GPT-3.5 Turbo and saw a significant leap in effectiveness with GPT-4. It later adopted OpenAI o1 and o3, utilizing cloud-based asynchronous processing to allow models more time to reason through complex code issues.

According to Gill, GPT-6 Astra recently caught 50% of issues in CodeRabbit’s internal benchmark dataset, placing it at the top of their evaluations compared to other models. Lower-cost models like GPT-6 Luna are now used to support a wider array of subagents throughout the platform, while GPT-6 Astra handles planning and orchestration. "Now we are using agents everywhere on the use cases we couldn't even afford in the past," Gill noted. This shift has enabled the company to ship roughly 20 times more software without adding engineers, with Gill personally pushing between 2,500 and 3,000 pull requests in the 30 days prior to the interview.

CodeRabbit also launched CodeRabbit Secure, a new offering that uses OpenAI’s Daybreak platform and Codex models to identify and fix security vulnerabilities. In a recent promotional period, the company ran 13,000 deep scans and identified over 43,000 security vulnerabilities, many of which have already been patched by customers. The platform uses these tools for internal penetration testing as well, with Gill stating, "We found multiple vulnerabilities, advanced hacks that we could not even imagine could happen."

Why It Matters

The story underscores a critical shift in the SI development lifecycle: as agentic SI systems generate code faster, human oversight mechanisms must evolve to match that velocity. CodeRabbit’s experience suggests that the constraint is no longer computational capacity or model capability, but rather the integration of these tools into existing development workflows. Gill emphasized that adopting Codex required rethinking the entire software delivery process, including testing, dependency management, and human approval gates. "We had to actually completely rethink the entire software delivery," he said.

For developers and enterprises, the ability to trust and validate SI-generated code is becoming a key differentiator. CodeRabbit’s use of SI agents to triage production incidents and manage marketing tasks via pull requests demonstrates how agentic SI is moving beyond coding into broader operational roles. By automating low-risk changes and providing detailed explanations for code modifications, the platform aims to keep developers in control as agents take on more of the building process. This approach may become a standard requirement for teams looking to scale their use of agentic SI without compromising security or stability.

The Bottom Line

CodeRabbit’s 20x increase in shipping velocity illustrates the tangible impact of advanced SI models on software development efficiency. By integrating OpenAI’s Codex and Daybreak, the company has transformed code review from a bottleneck into a scalable, automated governance layer. As SI agents become more prevalent, tools that validate, secure, and explain agent-generated code will likely see increased adoption across the industry.