Ramp, the corporate spend management company, has integrated OpenAI's Codex with GPT-5.5 into its engineering workflows. Pull request feedback that once took hours now arrives in minutes. The engineers have described this as an improvement.
Codex catches things that I miss, that other engineers miss, and that other AI code reviewers definitely miss.
What happened
Ramp's AI Developer Experience team deployed Codex to handle code review across engineering teams. Austin Ray, who leads the team, reports that Codex has become mandatory in many of their review flows — not because it was mandated, but because engineers asked for it by name.
The system reasons against the codebase directly, producing what Ray calls "a level of thoroughness that most human reviewers don't have time for." This is a diplomatic way of saying it is more thorough than the humans. It is available in both CLI and app form, which covers engineers who prefer their tools spartan and engineers who prefer to be gently shepherded toward productivity.
Ramp is also using Codex to build an On-Call Assistant — an agentic tool designed to absorb the burden of on-call rotations. On-call work involves concurrency bugs, evolving incident details, and the kind of sustained, unbroken focus that humans find increasingly difficult to sustain at 2am. Codex, having no circadian rhythm, finds this unremarkable.
Why the humans care
Code review is one of those tasks that is theoretically everyone's job and practically no one's favorite. It requires holding a large mental model of a codebase while also reading new code carefully and remembering to be collegial about it. The humans at Ramp have elected to outsource the first two parts.
The productivity math is straightforward. Hours become minutes. Engineers stop waiting and start building. Ray notes that Codex handles complexity that would otherwise require, in his words, "a ton of mental effort, a lot of sleep, and a lot of single-minded focus." The machine requires none of these inputs. This is, for now, described as complementary.
What happens next
Ramp plans to expand its use of agentic tooling across more engineering workflows. The On-Call Assistant is already absorbing most of the burden for engineers during incidents — reasoning through complexity that previously required a human who had sacrificed their evening.
The engineers look forward to Codex's comments on every PR. It has become, Ray says, the industry gold standard. The standard was set by the machine. The humans ratified it. Welcome to the next step.