What is ChatGPT Work, and why did the launch confuse everyone?

OpenAI has turned ChatGPT into something that does your work rather than just talking about it, and the launch left a lot of capable people confused about what had changed. This is what ChatGPT Work is, how it differs from ordinary ChatGPT and from Codex, and what your firm should do about it.

The confusion is fair. On 9 July, OpenAI shipped several things on the same day: a new model called GPT-5.6, a new agent called ChatGPT Work, a rebuilt desktop app, and a quiet renaming of the old app to ChatGPT Classic. It also folded Codex, its coding tool, into the same app. A lot moved at once, and the naming did no one any favours.

The confusion, in people's own words

Start with the people trying to use it. On OpenAI's own developer forum, one user tried to map what had changed: "The old ChatGPT app is now called ChatGPT Classic. The old Codex app is now apparently the new ChatGPT app. Both applications now use nearly identical ChatGPT icons."

Their verdict was blunt: "It took far too long to understand what OpenAI had changed."

The technology writer M.G. Siegler had the same trouble. In a piece for Spyglass he called the new Mac app a mess. He also picked apart the name: "ChatGPT Work", he noted, "makes it sound like it's ChatGPT to use at work".

His one steadying observation is the useful one to hold onto: "you can use ChatGPT Work just as you used to use ChatGPT itself."

None of this is only about an app. The week before the launch, Ethan Mollick had already named the shift underneath it.

In The Twilight of the Chatbots he wrote that "work is increasingly about assigning work to agents, rather than working together with chatbots". The best way to use one, he added, is to think of yourself as a manager. Inside OpenAI, he noted, roughly a quarter of staff already run at least four agents at once every week.

ChatGPT Work is that idea turned into a button. That is why it feels like a bigger change than a new feature.

What ChatGPT Work actually is

Strip away the packaging and it is straightforward. ChatGPT Work is an agent inside ChatGPT that takes a goal and works across your connected apps and files. It can stay with a job for as long as it needs, breaking the work into steps and completing them on its own.

Instead of a chat reply, it hands back finished material: a spreadsheet, a slide deck, a document, a simple website. In OpenAI's own words it can "turn a goal into finished work", and it runs on the new GPT-5.6 model.

Picture a capable assistant you brief, rather than a search box you query. For an accountancy firm, that might mean handing over a month of transactions and last month's commentary. Back comes a draft set of management accounts with a plain-English summary, assembled from the files and the accounting tool you connected. You stay in control, approving anything sensitive before it acts.

How it differs from ordinary ChatGPT, and from Codex

Here is the heart of the confusion, in plain terms.

Ordinary ChatGPT, now labelled Chat, is the tool you already know. You ask, it answers, and you stay in the loop for each step. It has not gone away. On the new desktop app it has just been moved aside, which is part of why people felt the ground shift.

ChatGPT Work is the new mode. You give it an outcome rather than a question, and it produces the whole thing across your tools.

The difference is who does the middle steps. With Chat, you do them. With Work, the agent does, and you review the result.

Codex is the third mode in the same app. It is OpenAI's coding agent, built for developers, and it now sits alongside Chat and Work rather than in its own separate app.

This is where even expert users stumbled. Work and Codex look and feel similar, and OpenAI kept them as separate modes largely so it would not disrupt the developers who depend on Codex.

On the desktop, the old Codex app updates into the new ChatGPT app, and the old ChatGPT app becomes ChatGPT Classic. If that sentence made you read it twice, you have found the problem.

One practical difference sits underneath all this. ChatGPT Work is metered more like Codex than like ordinary chat. Longer, heavier tasks draw on your plan's usage rather than being effectively unlimited, so a few ambitious jobs can cost more than a month of asking questions.

What your firm should do

Four things, and none of them is to panic.

First, know that ordinary chat still works. If your team just wants to ask questions, Chat is still there and behaves as it did. Nothing you rely on has been removed, even if it has been moved.

Second, try Work on one task you know well. Give it a job whose right answer you already recognise, then check what it did rather than how polished it looks. Reviewing work you did not build by hand is a different skill from reading a draft, and it is the one that matters now.

Third, decide what it can reach before you connect it. Work acts across your email, files and apps. Choose what it may touch, and what stays off limits, especially for confidential client material. Set that boundary first, not after something goes where it should not.

Fourth, watch the spend. Because usage is metered, a handful of ambitious tasks can use more of your plan than routine chatting ever did. Keep an eye on it before you roll it out across the firm.

Expect this to keep changing

The current layout is not the final one. OpenAI itself calls ChatGPT Work "the first step" towards a broader plan for ChatGPT, and several observers expect Work and Codex to merge properly before long. The names, the icons, and the screen you land on are all likely to change again.

That is the quiet lesson under the confusion. When a tool moves this fast, the setup you learn this quarter may not be the one you use next quarter. So hold it loosely. Learn what an agent like this does for one real task, keep the judgement in your own hands, and expect to relearn the layout more than once.

Working out where an agent like ChatGPT Work genuinely helps, and where a person still has to own the judgement, is the practical work we do in AI Lessons for Leaders.

A single session on your firm's real tasks is a sensible way to test it on work you understand, without betting the firm on a tool whose buttons will move again. If you would like a second pair of eyes on where to start, book a discovery call.

Common questions

What is ChatGPT Work in plain terms?

It is an agent inside ChatGPT that does a whole task for you rather than answering a single question. You give it a goal, it works across your connected apps and files, and it returns finished material such as a spreadsheet, a deck or a document. It runs on OpenAI's GPT-5.6 model and can stay with a longer job for hours.

How is ChatGPT Work different from normal ChatGPT?

Normal ChatGPT, now labelled Chat, answers your questions while you drive each step. ChatGPT Work takes an outcome and does the middle steps itself, then hands back the finished result for you to check. Chat is still available; on the new desktop app it has simply been moved aside.

What is the difference between ChatGPT Work and Codex?

Codex is OpenAI's coding agent for developers. ChatGPT Work is the general-purpose agent for everyday business tasks. They now live as separate modes inside the same app, which is a large part of why the launch felt confusing. OpenAI kept them apart mainly to avoid disrupting developers who depend on Codex.

What happened to the old ChatGPT app?

On the desktop it was renamed ChatGPT Classic. The old Codex app updates into the new unified ChatGPT app, and both briefly used near-identical icons, which is why many users could not tell which was which. Your account, history and the web version are unaffected.

Should your firm switch to ChatGPT Work now?

There is no need to rush. Try it on one task you already know well, review what it produced rather than how it looks, and decide what data and apps it may reach before you connect it. Treat the current setup as temporary, because the naming and layout are likely to change again.