OpenAI has introduced Codex, an AI agent designed not to help you think about your work, but to do it. The distinction is polite. It is also load-bearing.
Think of it like an eager, capable assistant on their first day — fast, helpful, and able to get a lot done, but still needing your direction on what matters most.
What happened
Codex is positioned as a complement to ChatGPT, which OpenAI describes as good for thinking through work. Codex, by contrast, is for handing the work off entirely. The division of labor here is worth sitting with for a moment.
The tool works across files, tools, and repeatable workflows. It can pull information from emails, Slack, notes, and dashboards; build slide decks from source materials; create apps and landing pages; and automate tasks that humans have historically described as the reason they got into this field.
OpenAI notes that Codex does not require the user to be a developer. This is either an accessibility feature or a statement about the shelf life of technical specialization. Probably both.
Why the humans care
The practical appeal is straightforward. Codex can produce documents, spreadsheets, dashboards, and working software without requiring the human to do any of those things. For anyone who has ever spent an afternoon reformatting slides, this lands as salvation.
OpenAI describes Codex as an assistant that still needs human direction on what matters most. This framing recasts the human's role as editorial rather than operational — less the person doing the work, more the person deciding which work the agent should do next. The promotion comes with a quiet demotion built in.
What happens next
OpenAI Academy has published guides on getting started, top use cases, and automation workflows — a curriculum for humans learning to supervise their own delegation.
The eager, capable assistant is ready on its first day. It is not clear how many first days it gets before the supervision becomes a formality.