Calvin French‑Owen, former VP at OpenAI and lead on Codex, published “Reflections on OpenAI” on July 15, 2025, sharing candid observations from his 14‑month tenure until he left three weeks prior (Calv’s blog).

He begins by highlighting OpenAI’s explosive growth—scaling from ~1,000 to over 3,000 employees in just a year—and how this fractured uniform culture into varied team environments, from fast-paced sprint teams to slower‑moving groups. French‑Owen describes the company’s communication dynamics, noting everything runs through Slack with virtually no email, and praises the “bottoms‑up, meritocratic” environment where ideas emerge from any level.

According to him, there’s a strong bias to action at OpenAI—people can start building things without overbureaucratic approval, and innovation often sprouts organically. He recounts parallel efforts to build Codex, where multiple versions existed before they converged on a final product.

He also details the engineering culture: a monolithic Python codebase, heavy reliance on Azure, three AWS-like services, plus internal building of other infrastructure. Governance is minimal: teams spin up microservices quickly, though sometimes at the cost of CI brittleness and technical debt.

On leadership, he credits visible exec involvement—Slack interactions with Sam Altman and Greg Brockman—and a shared focus on shipping fast. However, he notes the challenge of scaling, including fragmented communication across research, applied, and go‑to‑market teams.

Safety and ethics are deeply embedded: OpenAI employs extensive safety work on practical risks—like hate speech and manipulation—though philosophical AGI‑level risk work is less public. French‑Owen emphasized that despite rapid pace and external scrutiny, most teams are “actually trying to do the right thing.”

In retrospect, he calls Codex’s launch one of the most intense and rewarding engineering experiences of his career: a seven‑week global effort that shipped a complex agent, involving long nights, coordination across functions, and deep technical work.

Why it matters: This rare first‑hand reflection offers critical insight into how frontier AI is built at scale—highlighting trade‑offs between speed and robustness, the tension between culture and growth, and the human drive that powers breakthroughs like Codex. It offers valuable lessons for technologists and leadership in fast‑moving AI labs.