How PMs Ship 100K Lines of Code at OpenAI with Ryan Lopopolo, Member of Technical Staff

Software engineering is undergoing a fundamental shift where code is increasingly treated as a liability rather than an asset, necessitating a move toward fully agentic development. Ryan Lopopolo, a lead on OpenAI’s Frontier team, advocates for building a "harness"—a repository-level infrastructure that injects context, documentation, and quality guardrails directly into AI agents. This approach enables non-engineers, including product managers and designers, to contribute directly to the codebase by utilizing prompt engineering and automated validation loops. By treating the codebase as a "token factory" and saturating models with high-quality context, teams can achieve massive parallelism, reducing the need for manual, synchronous coding. This model replaces traditional, friction-heavy handoffs with a unified, agent-driven workflow that accelerates product experimentation and delivery, ultimately allowing organizations to operate at a significantly higher velocity.