Ramp uses Codex to speed code review and agent tooling

Ramp says its engineers are using Codex with GPT-5.5 to speed up code review and build internal agentic tooling, including an internal On-Call Assistant. According to Austin Ray, who leads AI Developer Experience at Ramp, the tool gives teams substantive pull request feedback in minutes rather than hours and reduces manual work that would otherwise slow engineers down.

Ray said Codex has become a regular part of Ramp’s development process. He described Codex code review as the “industry gold standard” and said engineers ask for it by name. In some workflows, he said, it is now a mandatory step before code moves forward.

The core reason, according to Ray, is that Codex reasons deeply about the codebase. He said that lets it catch issues that he and other engineers miss, along with problems that other AI code reviewers also miss. That depth matters at Ramp, where the company’s software includes a large amount of business logic and domain knowledge.

Ramp says the review speed difference is significant. Engineers who previously waited hours for a first review can now get meaningful feedback in minutes. Ray said that the system offers a level of thoroughness that most human reviewers do not have time to match.

Codex is also being used to support internal tool development. Ray said he is using it while building On-Call Assistant, an agentic tool designed to take on much of the work Ramp engineers face during on-call rotations. He said on-call work is difficult because engineers have to keep many things in context, including concurrency bugs, outside and internal events, and long-running incident investigations that change as new details emerge.

According to Ray, Codex with GPT-5.5 helps him manage that complexity. He said it can handle Ramp’s large product surface area with ease and that it makes him more confident about the changes he ships. That has also made the On-Call Assistant faster to build.

Ray said engineers can use Codex in different ways depending on how they like to work. Those who prefer working close to the command line can use the CLI, while the Codex app provides visual cues, utilities, and other features for people who want them. Ray said he normally prefers the CLI but was drawn to the app because it guided him toward higher productivity.

For company leaders evaluating AI tools, Ray said the important question is whether a tool actually changes how people ship code or is just a demo. His advice is to have engineers install it, walk them through a strong first session, and build trust through repeated use.

He also said the feedback loop with the Codex team matters. Ramp works directly with the vendor when issues come up, and Ray said that relationship has helped the company make strong progress.

Looking ahead, Ray said tools like Codex point to a shift in the engineering role. In his view, engineers are becoming orchestrators who direct AI tools, decide when to trust them, and know when to push back. At Ramp, he said, the best engineers are learning that skill fastest.