How Ramp engineers accelerate code review with Codex

Ramp Integrates Codex with GPT-5.5 to Accelerate Code Review and Build Internal Agents

Engineers at fintech company Ramp are using OpenAI's Codex, powered by the GPT-5.5 model, to significantly reduce the time required for code review and to develop sophisticated internal tools. The implementation moves beyond basic code completion, providing substantive feedback on pull requests in minutes rather than hours. This application demonstrates a maturing use case for advanced AI in core software development cycles, focusing on deep code analysis to improve both development velocity and final code quality.

According to Austin Ray, who leads the AI Developer Experience (AI DevEx) team at Ramp, the tool's primary advantage is its ability to reason deeply against the company's extensive codebase, catching complex issues that both human reviewers and other AI tools often miss. The platform is being used for more than just code validation; Ray's team is also leveraging Codex to build an internal agent called 'On-Call Assistant'. This tool is designed to manage the high cognitive load of on-call engineering rotations by handling complex business logic, concurrency bugs, and long-running incident investigations.

• Reduces first code review wait times from hours to minutes.
• Provides deep reasoning against the entire codebase, not just isolated snippets.
• Supports development of complex internal agents like the 'On-Call Assistant'.
• Offers flexible developer experience through both a CLI and a dedicated app.
• Mandatory in many of Ramp's code review workflows.

The successful deployment at Ramp signals a potential evolution in the role of software engineers, shifting them from pure producers of code to orchestrators of AI development tools. Ray suggests the critical skill is now knowing how to direct, trust, and challenge AI assistants like Codex. For the broader market, this sets a higher standard for developer AI tools, prioritizing deep, contextual reasoning and seamless workflow integration over surface-level features. The case also highlights the value of a direct feedback loop between enterprise users and AI vendors like OpenAI for refining these complex systems for practical, real-world deployment.

Strategic Takeaway: Ramp's adoption of Codex demonstrates that the enterprise value of developer AI lies in its ability to deeply reason about proprietary, complex codebases within existing workflows. Success is less about autonomous code generation and more about creating a trusted, collaborative tool that augments senior engineering tasks like code review and incident management, driven by a tight vendor-client feedback loop.