How Ramp Uses Codex to Replace Hours of Code Review

Key insights

• Ramp uses Codex not just for writing code but as an on-call assistant, showing how AI coding tools are expanding into operational workflows.
• GPT 5.4 handles complexity that would require 'a ton of mental effort, a lot of sleep, and a lot of single-minded focus' from a human engineer.
• Code review at Ramp went from hours to minutes: one of the clearest before/after productivity metrics in enterprise AI adoption.

In Brief

Ramp calls OpenAI's Codex code review "industry gold standard". Engineers used to wait hours for a first review. Now Codex (an AI tool that can write, review, and fix code) gives them substantive feedback in minutes. Austin Ray, who leads the team that builds AI tools for other developers at Ramp, shared how his team relies on Codex daily and is now building something bigger: an AI-powered on-call assistant that takes the operational burden off engineers entirely.

From hours to minutes

Code review is one of the biggest bottlenecks in software development. Before a change can ship, another engineer has to read the code, catch mistakes, and suggest improvements. That takes time, often hours in a busy team.

Ramp found a way to shrink that wait dramatically.

"Engineers used to wait hours for a first review. Now, Codex gives them substantive feedback in minutes."

The result: engineers spend less time waiting and more time building. Codex runs as an AI agent, a system that takes actions on its own rather than just answering questions. It works through a codebase the way a human reviewer would, just much faster.

The on-call assistant

Ramp's AI DevX team builds tools that make other developers more productive. Their current project takes that mission further than code review.

"The current project I'm working on is on-call assistant, an AI-driven on-call helper to take most of the burden off of our engineers."

On-call is the rotation where engineers must be available to fix problems the moment they appear in production, meaning the live system real users depend on. Think of it like a doctor on call at a hospital: stressful, sleep-disrupting, and requiring deep knowledge of the system. Building an AI that can handle most of those incidents automatically is a serious undertaking.

Codex is the foundation Austin Ray is building it on.

Handling complexity that exhausts humans

Why Codex specifically? Ray points to something that goes beyond speed: the ability to reason through genuinely difficult problems without losing focus.

"Codex with GPT 5.4 is incredibly adept at dealing with complexity in a way that would take me a ton of mental effort, a lot of sleep, and a lot of single-minded focus."

GPT 5.4 is the OpenAI model that powers Codex. Ray's point is that some engineering problems aren't just time-consuming — they're mentally exhausting. Tracing a bug through layers of interconnected systems, holding a large mental map while testing hypotheses, staying focused for hours. These are things humans do poorly when tired.

Codex with GPT 5.4 just sort of handles it like it's nothing.

What this shift looks like

AI coding tools started as autocomplete: suggestions while you type. Then they became pair programmers. Now they're reviewing code, responding to production incidents, and working through complex problems on their own.

Ramp's experience shows the developer workflow shifting from wait for a human to ship with AI support. Hours to minutes on code review. An AI handling on-call incidents that used to cost engineers sleep. These aren't demos. They're already running in production.

Glossary

Sources and resources