"Just Ask Codex": Inside OpenAI's Own Working Culture | Romain Huet, OpenAI

Codex has evolved from a developer-focused IDE tool into a powerful agentic command center, enabling users to delegate complex, multi-hour, or even multi-day tasks autonomously. This transformation relies on the synergy between high-intelligence models like GPT 5.5 and a robust "harness" layer, which provides agents with necessary context, tool access, and self-correction capabilities. By moving beyond simple chat-based interactions to goal-oriented execution, users can now assign ambitious objectives—such as large-scale code refactoring or bug remediation—that the AI pursues through iterative testing and environment navigation. Security remains paramount, with built-in sandboxing and explicit user permissions ensuring safe operation across local file systems and applications. This shift marks a fundamental change in how work is structured, moving from manual task execution to managing intelligent agents that handle the heavy lifting of daily professional workflows.