The End of Git? OpenAI’s CTO on the Future of Coding

As coding agents speed up development, bottlenecks are shifting to reviews, CI and repository infrastructure—and voice may widen ChatGPT’s reach.

Coding agents are changing how software gets built inside OpenAI, Vijay Raji, the company’s applications CTO, says. Since February 2026, stronger models, Codex and parallel agents have driven a sharp increase in pull requests. Faster code generation, however, moves the bottleneck to reviews, testing, CI/CD and repository infrastructure.

OpenAI built an internal system for its monorepo. Raji argues that Git’s traditional workflow assumes developers work offline and push later, while agents operate continuously. At that scale, checkout throughput and caching become major concerns.

The conversation also turns to ChatGPT’s next billion users. Raji sees voice as a key interface for people who are less comfortable typing or using English. He also says a roughly 50-person advertising team reached a billion-dollar annualized run rate within months by using models across the ad pipeline.

His test for an AI-native product is straightforward: each stronger model should make the product more valuable, not obsolete.

Source: Nate B Jones’s interview with OpenAI applications CTO Vijay Raji.

Source