Descriptions:
At The Pragmatic Summit, two OpenAI engineering leaders — VJ and Tibo, who leads the Codex team with 33 direct reports — describe how software development has already fundamentally changed inside OpenAI. The shift is not incremental: engineers are routinely consuming hundreds of billions of tokens per week through Codex, and the team has shipped Codex Box internally — a system that reserves cloud-based development machines, allowing engineers to queue up prompts, close their laptops, attend meetings, and return to completed work. The framing is explicit: Codex has evolved from tool to extension to agent, and both speakers expect engineers to soon name their agents and treat them as teammates in a formal sense.
Tibo describes the Codex team’s structure as deliberately flat, with the team reinventing its own workflows on a near-weekly basis as bottlenecks shift. Code generation is largely solved; the current frontier is understanding user needs faster — synthesizing signals from Twitter, Reddit, and other surfaces using agents to inform product strategy. Onboarding itself is now largely handled by Codex: new engineers navigate the codebase by asking it questions, receive daily summaries, and are guided by the teammates who most recently went through the same process.
VJ addresses the widely debated question about junior engineers directly: OpenAI is actively hiring new graduates and plans an internship cohort of approximately 100 people in the coming summer, operating on the conviction that AI-native engineers entering the workforce will leverage these tools from day one in ways that make fears about AI displacing early-career developers largely unfounded. The conversation offers a rare first-person account of what AI-assisted engineering looks like at production scale inside one of the field’s leading organizations.
📺 Source: The Pragmatic Engineer · Published February 23, 2026
🏷️ Format: Interview
