OpenAI Launches ChatGPT Work — an AI Agent for Your Job

OpenAI today launched ChatGPT Work, a new autonomous agent that lives inside ChatGPT and handles complex, multi-step tasks on your behalf. Instead of answering one question at a time, Work takes a goal, gathers context across your connected apps, breaks it into steps, and returns finished deliverables — spreadsheets, slide decks, documents, and even interactive web apps.

What It Does

ChatGPT Work is designed for the kind of tasks that normally eat up your afternoon. Give it a brief like “research our top five competitors and build a comparison deck,” and it will browse the web, pull data, and produce a finished presentation. It can access local files, use a built-in browser to visit websites and online tools, and stay with complex projects for hours without needing you to babysit the process.

The key shift is in what it returns. Where regular ChatGPT gives you text in a chat window, Work delivers actual artifacts you can use immediately — no copy-pasting into another app.

How It Fits Together

Work arrives alongside two other major changes from OpenAI. First, GPT-5.6 — the model family OpenAI previewed in late June — is now generally available to all ChatGPT and API users in three tiers: Sol (flagship), Terra (balanced), and Luna (fast and affordable). ChatGPT Work runs on GPT-5.6 under the hood.

Second, OpenAI is consolidating its desktop apps. The standalone Codex app is being folded into ChatGPT, and the Atlas browser is being sunsetted. Work effectively replaces both as a single agent that handles coding, research, and general productivity tasks in one place.

Who Gets It

ChatGPT Work is rolling out today on web and mobile, starting with Pro, Enterprise, and Edu subscribers. Plus and Business users can expect access within the next few days.

What This Means for You

If you use ChatGPT for work, this is a meaningful upgrade. Rather than prompting your way through a task step by step, you can now describe the end result you want and let the agent figure out the path. It is a shift from “AI as a chat tool” to “AI as a co-worker” — and it puts OpenAI in direct competition with tools like Claude Code and Google’s Project Mariner that are also racing to build reliable AI agents.

For now, the practical advice is simple: try giving it a real task you would normally spend an hour on and see what it produces. That will tell you more than any benchmark.