ChatGPT might not be available in the future.

OpenAI's super app has finally taken shape.

It's still named ChatGPT, but it's about to have a new addition: Codex.

Moreover, ChatGPT is basically just a shell now.

At the Intelligence at Work press conference just held by OpenAI, the official announced that Codex will be integrated into ChatGPT within the next few weeks.

When I first heard this news, I was a bit slow to react. Isn't there always a Codex option in ChatGPT?

I just clicked on it today and found that it was a "fake" option, which was meant to lead you to download the Codex app. You can't use it directly in ChatGPT.

So, after the integration, chatting and working can be done in one app, and you don't have to switch between apps anymore.

But do you think that's the end of the story?

At least from OpenAI's perspective, Codex might just be the beginning.

Through this press conference, we finally got a clear picture of all the ideas about the super app.

Actually, Codex is just the first step

The press conference lasted for a whole hour and was packed with information. But in summary, there are three core points.

1. ChatGPT is going to merge with Codex because the times have changed.

2. Codex will continue to be upgraded, so three major updates are introduced.

3. Codex is catching up with Claude Code, and the "efficiency" of GPT - 5.5 is a key factor.

Let me give you a quick overview:

First, there's the rather unexpected merger. Why did ChatGPT choose to merge with Codex now?

One reason is that Codex, which was launched to catch up with Claude Code, can now prove its worth with results.

The weekly active users of Codex have exceeded 5 million, a six - fold increase since its desktop launch in February.

More importantly, 20% of its users are not programmers at all, but knowledge workers such as analysts, designers, and investment bankers. Moreover, the growth rate of this group is three times that of developers.

This shows that Codex is making a strong breakthrough across different user groups.

Regarding the revenue issue that OpenAI values highly (they followed Claude Code because of its profitability), Codex itself revealed that the revenue from enterprises currently accounts for 40% of OpenAI's overall revenue, and this proportion will reach 50% by the end of this year.

To put it simply, Codex now has a rapid user growth, a wide user base, and it's actually making money for OpenAI.

It seems that everything is telling OpenAI that it's time to consider the next step.

And the direction for the next step is quite clear at least for now - from chat to agent, from chatting to execution.

Alexander Embiricos, the product leader at OpenAI, said this at the press conference:

You may not work 24/7, but your agent in the cloud will.

In just one line, he revealed the ultimate charm of a super app. In the future, users really won't have to switch between different apps. You give an instruction, ChatGPT can understand it, and then Codex can call various agents to execute it for you.

So this integration is actually the first step for OpenAI towards a super app.

You may ask what the next step is. The answer might be what you've already guessed - the browser.

The browser is the last piece of the integration puzzle. It fills in the entry for AI to enter the web world, allowing users to simply give instructions in ChatGPT, and Codex can automatically complete tasks such as searching, operating the backend, and handling work in the browser, which originally required manual clicks.

So, with ChatGPT + Codex + Atlas browser, OpenAI finally has the hope of achieving full - scenario execution from chatting through a single app.

But for now, let's just focus on Codex.

To give Codex more "soldiers" to do the work, OpenAI has introduced three more updates in one go:

Agent plugins for six major positions, covering data analysis, sales, creative production, product design, equity investment, and investment banking. It directly integrates 62 enterprise applications such as Snowflake, Figma, and Salesforce, and has 110 built - in automation skills.

Annotations function, which allows you to select and modify directly on the original document without having to regenerate the whole thing. Developers have already been using this function to modify code, Markdown files, and websites generated by Codex.

Now, it has been extended to content creation scenarios such as documents, spreadsheets, and PPTs.

Sites function, which enables Codex to turn your work results into an interactive web application with one click and generate a URL for direct sharing with the team.

This function seems quite practical. In the past, when developing a project, non - technical people could only vaguely describe their requirements. Now, everyone can generate a clear and intuitive visual demo, which will obviously significantly reduce the team communication cost.

Currently, this function is available for preview to Business and Enterprise customers.

See, OpenAI is betting that more and more ordinary people will shift from dialogue to execution in the future.

Some people may ask, why doesn't Codex absorb ChatGPT?

It can only be said that on the surface, it seems like ChatGPT is absorbing Codex, but when you put several details together, the direction might be the opposite.

In the organizational structure adjustment in May, Thibault Sottiaux, who built up Codex, was promoted to the head of core products and platforms, in charge of consumer, enterprise, and developer lines. Nick Turley, who made ChatGPT reach 900 million weekly active users, was transferred to the enterprise product line.

So now, it's the people from Codex who are in charge of ChatGPT, not the other way around.

Moreover, the wording in Brockman's internal memo is also worth pondering -

He didn't write "upgrade ChatGPT", but "invest in a unified agentic platform".

So this merger is more like a route switch than a product integration: from dialogue to execution, from chat to agent.

ChatGPT provides a traffic base of one billion users, while Codex provides the growth engine and future narrative.

The shell is ChatGPT's, but the soul is Codex's.

However, Codex's development was ultimately forced by its competitor, Anthropic.

Even GPT - 5.5 mentioned in the press conference can almost be regarded as a product of this pressure.

Codex was forced out by Anthropic

Looking back, if it weren't for Anthropic, Codex might still be an unremarkable auxiliary function in ChatGPT today.

Let's rewind the timeline to when the preview version of Claude Code was released last year.

Claude Code was a great success as soon as it was launched. Its annualized revenue exceeded $2.5 billion in February this year. It only took six months to reach $1 billion in revenue from its launch, making it one of the fastest - growing products in the history of commercial software.

About 4% of global public GitHub submissions are from it, and users spend an average of 20 hours per week using it.

It was the success of Claude Code that made OpenAI realize internally that it had fallen behind in the coding field.

What to do when falling behind? Catch up.

In the past, the default narrative in the AI industry was that Anthropic was following in OpenAI's footsteps - GPT came out first, and Claude followed; ChatGPT became popular first, and Claude made a counterpart.

But on the Code front, the script has been completely rewritten.

Claude Code was launched in February last year, and OpenAI didn't launch the counterpart Codex until two months later; Anthropic developed the desktop version first, and OpenAI followed suit; Anthropic launched Cowork for knowledge workers, and OpenAI later chased with Plugins.

This time, it's OpenAI following Anthropic's lead.

However, the catching - up speed is really fast. Looking at the timeline, in 14 months, it went from zero to 5 million weekly active users.

In April last year, the preview of Codex Cloud was launched;

In February this year, the desktop app and the dedicated model GPT - 5.3 - Codex were launched;

In March, the merger of the super app was announced, and the Python tool provider Astral was acquired;

In April, ChatGPT Pro at $100 per month was launched to attract developers;

In May, the mobile version was launched;

In June, it was officially integrated into ChatGPT.

It can be said that almost every step has the shadow of Claude Code.

To be honest, Codex's comeback doesn't rely on competing head - on in code quality.

After all, the blind - test data shows that Claude Code still has a 67% win - rate, and most programmers admit that the code it writes is more solid.

But unfortunately, Claude Code has very strict usage limits and is more expensive than Codex.

A discussion thread on Reddit with more than 500 developers participating boiled down to a consensus:

Claude Code is smarter, but its limits are too strict to be used daily; Codex is a bit less capable but can really be used effectively.

One model is smarter but often gets stuck in the middle of tasks and is more expensive, while the other is slightly less smart but more affordable.

Which one would you choose? (doge)

Interestingly, Alexander, the product leader at OpenAI, also said this golden line at the press conference:

The thing GPT - 5.5 hates the most is wasting tokens.

By using Codex with GPT - 5.5, you can get the same - quality output with only one - third of the tokens.

In short, on the road of catching up with Claude Code, Codex is making "more intelligence with fewer tokens" its new principle.

There's no denying that the AI programming field is currently in a situation of a two - horse race.

One represents the strongest code - writing ability, and the other represents the strongest productization ability.

But if you look at the long - term, you'll find that what Claude Code and Codex are competing for is no longer just the programming tool market.

As more and more ordinary people start to use agents as their daily work partners, the entry point is actually the most important thing.

And this gives Chinese players new opportunities - because in the agent era, what determines the outcome may not just be the model's ability, but also the scenario, ecosystem, application connection ability, and the understanding of the local enterprise workflow.

In fact, domestic big tech companies have gradually entered the arena.

Some of them choose to target the developer entry point, some target the internal enterprise office entry point, and some target the full - process intelligent agent platform.

But no matter which path they take, the goal is basically the same - to seize the core entry point in the agent era.

Everyone knows very well that the real urgent task is how to take advantage of the time window of the local scenario to build a real competitive barrier at the agent infrastructure level.

So, let's see who can be the first to stand out and create the Chinese version of Claude Code and Codex?

And more importantly, who can become the new super entry point in the domestic agent era?

One More Thing

Let's go back to this merger event.

Judging from Codex's current development trend and the industry environment, I'd like to boldly say:

Instead of continuing to call the merged app ChatGPT, it might be better to call it Codex directly.

Think about it. ChatGPT sounds like an old product from the "chat" era, but the direction OpenAI is betting on is no longer chat.

The product is managed by people from Codex, the internal memo mentions an agentic platform, the fastest - growing scenarios are execution - related, and none of the new features are related to chatting.