Wiki › People & organisations

OpenAI Codex

OpenAI coding-agent product and delegated-work surface whose harness primitives include automations, worktrees, connectors, browser/computer use, and cross-app coordination.

OpenAI Codex is an AI coding-agent product and harness surface. In the wiki it appears first as one of several agent harnesses alongside Claude Code, Cursor, Aider, and Cline: the model matters, but the surrounding app, tools, state, context, worktree handling, connectors, and feedback loops strongly shape what the user experiences.¹

Addy Osmani's loop-engineering source treats Codex as part of the productization of Agent Loops: automations, /goal-style continuation, worktree isolation, skills, connectors, sub-agents, and durable state are becoming built-in product primitives rather than custom shell scripts around a model.²

Andrew Ambrosino's Lenny's Podcast interview extends the frame from coding agent to work home base. He says the Codex desktop app began as a right-sized developer surface rather than an IDE, but internal dogfooding showed marketing, comms, finance, legal, and other non-engineering teams using it even when the product was still developer-shaped. The lesson was that users were responding to a delegated-work surface, not merely a code editor.³

The product vision in the interview is not that every task must happen inside one rectangle. Codex can coordinate with other applications—browsers, Slack, Excel, Premiere Pro, and local tools—when those are the right surfaces. In this sense Codex sits between Harness Engineering and Personal Agents: it is a tool for coding, but also a general environment for starting, tracking, automating, and supervising work.³