- Sea Limited disclosed it has rolled out OpenAI‘s Codex across its engineering organization, with 87% of users active weekly per OpenAI internal data.
- Among Sea developers who rated Codex 4 or 5 out of 5, 73% said they would recommend it to colleagues.
- David Chen, Sea co-founder and Chief Product Officer of Shopee, framed the deployment as a structural shift in how engineering teams handle complexity.
- The interview was published May 14, 2026 as part of OpenAI’s Executive Function customer-interview series.
What Happened
OpenAI published a new entry in its Executive Function customer-interview series on May 14, 2026, featuring Sea Limited co-founder David Chen explaining why the Singapore-based conglomerate has rolled out Codex across its developer organization. The case study, posted at OpenAI’s blog, discloses two metrics OpenAI rarely shares publicly: 87 percent of Sea’s Codex users are weekly active users, and among developers who rated Codex 4 or 5 out of 5, 73 percent said they would recommend it to colleagues. Chen serves both as a Sea co-founder and as Chief Product Officer of Shopee, the conglomerate’s e-commerce arm.
Why It Matters
Sea’s three operating businesses — e-commerce arm Shopee, gaming subsidiary Garena, and digital financial services unit SeaMoney — together employ thousands of engineers across Southeast Asia, making the rollout one of the largest enterprise Codex deployments OpenAI has publicly disclosed. The 87 percent weekly-active number is materially higher than typical enterprise SaaS adoption benchmarks, where sub-50 percent weekly active rates are common even for well-received tools. The case study also lands the same week OpenAI integrated Codex into the ChatGPT mobile app and OpenAI CRO Denise Dresser told Bloomberg that enterprise revenue now accounts for 40 percent of the company’s top line, on track for 50 percent by year-end. Sea is precisely the type of named reference customer OpenAI uses to validate that 40-percent enterprise figure.
Technical Details
Chen described Codex’s distinguishing capability as “deep contextual awareness of our large and disparate codebases.” In Sea’s microservices architecture, he said, “the friction is not typing syntax. It is tracing dependencies, understanding legacy logic, and maintaining reliability under peak loads.” Chen characterized Codex as “a localised knowledge engine” that reduces the time engineers spend navigating unfamiliar services.
The deployment has moved beyond autocomplete-style assistance into what Chen called “integrated agentic workflows.” Specifically, AI agents now operate within Sea’s CI/CD pipelines — reasoning through product requirements, autonomously proposing test-driven implementations, surfacing edge cases in distributed systems, and accelerating debugging loops. Chen framed the discipline outcome explicitly: “we are also using it to drive engineering discipline. By allowing AI to rapidly prototype alternative implementations and generate exhaustive test coverage, we are moving faster and are systematically paying down technical debt and shipping more resilient systems.”
Who’s Affected
Sea’s engineering teams across Singapore, Indonesia, the Philippines, Vietnam, Thailand, and Brazil are the immediate population using Codex day-to-day. For OpenAI, Sea joins a small set of named enterprise reference customers used to demonstrate Codex penetration outside the United States and Western Europe — particularly important as the company markets to Asia-Pacific buyers. Competitors GitHub Copilot, Anthropic Claude Code, Cursor, and Windsurf are the named or implicit competitors for the slot Codex now occupies inside Sea’s stack. The case study also gives Southeast Asian and Indian enterprise buyers a regional reference point for adoption that did not previously exist in OpenAI’s publicly named customer roster.
What’s Next
OpenAI’s Executive Function series is expected to continue with additional customer interviews through 2026 as the company markets Codex against entrenched alternatives. Chen indicated Sea will continue evaluating where AI agents can take on whole work streams rather than augment individual engineer tasks, particularly in CI/CD pipelines where the company is already running agentic flows. OpenAI has not committed to publishing aggregate Codex usage data across its enterprise customer base, but the disclosure pattern in this case study — concrete percentages rather than vague descriptors — suggests more named-customer numbers will follow as competitive pressure from Anthropic and GitHub intensifies.
