▶Codex delivers significant productivity gains, with multiple sources citing dramatic reductions in development time and increases in output. For example, the Sora Android app was built in 28 days, a task that previously took weeks for multiple engineers can now be done by one in a week, and OpenAI engineers push 70% more pull requests.Feb 2026
▶Codex is heavily integrated into OpenAI's internal workflows, a practice often called 'dogfooding'. It is used by 95% of engineers daily, reviews 100% of pull requests, and is even used to write and review code for its own infrastructure management.Feb 2026
▶Codex is designed and used as an autonomous agent for complex, long-running tasks, not just as a simple code completion tool. Features like 'context compaction' enable it to work on jobs for extended periods, and it is preferred for difficult debugging scenarios like concurrency issues.Feb–Apr 2026
▶The adoption of Codex is fundamentally changing the software development workflow. Experts observe a rapid evolution away from traditional IDEs toward direct interaction with models like Codex, with some product managers now checking in code directly using such tools.Feb–Apr 2026
▶There is a contrast in the perceived user experience and 'personality' of Codex. While some describe it as a highly effective but 'dry' and 'uncommunicative' engineer, others like Andre Karpathy contrast this with the more 'teammate'-like feel of competitors like Anthropic's Claude.Feb–Apr 2026
▶While multiple sources agree that Codex is a powerful tool, there are differing views on the optimal usage pattern. Some claims highlight its effectiveness as a standalone agent solving complex bugs, while others describe emerging, more complex workflows involving fleets of 20-30 Codex agents managed by an orchestrator or using it in parallel with other models for 'peer review'.
▶The public perception of Codex has evolved. One claim notes that market perception shifted from favoring Anthropic's Claude Code to favoring Codex, associating the latter with a more rigorous, evaluation-focused culture. This suggests a period where it was not seen as the leading model, contrasting with its consistently deep internal use at OpenAI.Feb–Apr 2026
▶There's a subtle difference in describing the current state of the developer workflow shift. Bret Taylor describes a rapid evolution towards directly operating models like Codex as the final step, whereas Alexander Emberikos suggests most OpenAI developers have already abandoned traditional IDEs for tools like the Codex app, implying the new workflow is already established, at least internally.Apr 2026
Sign up free to see the full intelligence report
Get started free