Ramp engineers have integrated Codex with GPT‑5.5 into their development workflows to accelerate code review and to develop internal agentic tooling. According to Austin Ray, who leads AI DevEx at Ramp, the system now provides substantive pull request feedback within minutes instead of the hours engineers previously waited.
Faster, deeper code review
Ray describes Codex code review as able to reason deeply about the codebase, catching issues that both humans and other AI reviewers miss. That level of analysis has been folded into existing review flows, and some teams at Ramp now treat Codex feedback as a routine, expected step in pull requests.
Engineers continue to choose their preferred interfaces: those who work at the command line can keep that habit, while others use a Codex app that presents visual cues and utilities. Ray, a frequent CLI user, said the app helped guide workflows and handle complex reasoning that would otherwise demand concentrated mental effort.
On-Call Assistant and operational tooling
Ray is also using Codex to build On-Call Assistant, an agent intended to shoulder much of the operational burden during on-call rotations. Ramp’s description of on-call work highlights complex business logic, domain knowledge, concurrency bugs and long-running incident investigations with evolving details—conditions that make on-call duties strenuous for engineers.
Using Codex’s reasoning capabilities, the team reports faster development of the On-Call Assistant and greater confidence in shipped improvements. Ray said Codex handled Ramp’s large product surface area without apparent difficulty.
On leadership and adoption, Ray recommends hands-on demonstrations to build trust: have engineers install Codex, walk them through an initial session, and let them iterate. He also emphasises maintaining a feedback loop with the Codex team to resolve issues quickly. Ray characterised Codex as materially helping the team ship code faster and suggested that future engineering roles will focus more on directing AI tools than on writing every line of code.
Ramp’s internal use of Codex with GPT‑5.5 now informs both routine code review practices and the development path of agentic on-call tooling, and the team continues to work directly with the Codex developers on product feedback and improvements.
Original source: OpenAI News