ChatGPT and Codex Are Merging Into One Work Platform

OpenAI just made one of its clearest product moves yet: ChatGPT and Codex are no longer meant to feel like two separate products.

During its latest Intelligence at Work event, OpenAI signaled that Codex's core capabilities are moving deeper into the ChatGPT experience. The message is simple but important: ChatGPT remains the familiar conversational entry point, while Codex becomes the execution layer that can actually build, analyze, transform, and iterate on work.

In other words, ChatGPT is becoming less like a standalone chatbot and more like the front door to a working AI system.

That is a big shift.

For years, ChatGPT was where people asked questions, drafted text, summarized documents, and explored ideas. Codex, meanwhile, was mostly understood as a developer tool: something for writing code, reviewing pull requests, debugging projects, and automating software work.

Now OpenAI appears to be collapsing that boundary. Codex is still useful for developers, but it is also becoming a productivity agent for analysts, marketers, designers, sales teams, investors, bankers, operators, and other knowledge workers.

Why this matters

The important part is not just that Codex is being connected to ChatGPT. The important part is what OpenAI wants users to do with it.

OpenAI says Codex now has more than 5 million weekly active users, up more than 6x since the desktop app launched in February. Developers are still the largest user group, but non-developer knowledge workers now make up around 20% of Codex users and are growing more than three times as fast as developers.

That tells us something about where AI work tools are heading.

The next stage of AI adoption is not only about better answers in a chat window. It is about AI systems that can:

• connect to workplace context
• generate real deliverables
• update those deliverables after feedback
• turn plans into dashboards, reports, websites, and tools
• support multiple job functions without requiring every user to become a developer

That is the real reason the ChatGPT-Codex integration matters. OpenAI is trying to make Codex the engine behind everyday professional work, not just software engineering.

Codex is moving beyond coding

Codex started as a coding agent, and that history still matters. A system that can read files, reason through structure, edit code, run tasks, and iterate toward a result has a natural advantage over a chatbot that only writes text.

But OpenAI is now applying that same agentic pattern to broader knowledge work.

Inside a company, many tasks are not pure writing tasks. They are mixed tasks:

• analyze a dataset and make a report
• turn a strategy memo into a presentation
• build a dashboard from fragmented business context
• prepare a customer review workspace
• create a financial scenario planner
• convert a creative brief into assets and ad variations
• pull information from multiple tools and turn it into an action plan

These tasks require more than fluent language. They require execution.

That is where Codex fits. It can take a goal, use tools, produce artifacts, and revise them. When this ability becomes available through ChatGPT, the user experience becomes much easier: people can start with a conversation, then let Codex do the heavier work behind it.

The three big updates: plugins, Sites, and annotations

OpenAI's new Codex push is built around three major updates:

1. role-specific plugins
2. Sites
3. annotations

Together, these updates show how OpenAI wants Codex to work inside real teams.

1. Role-specific plugins

The first update is a set of plugins designed around specific job functions.

Instead of asking every user to configure a general-purpose agent from scratch, OpenAI is packaging Codex with role-specific workflows, instructions, integrations, and context. The initial set covers areas such as:

• data analysis
• creative production
• sales
• product design
• private equity
• investment banking

This is a practical move.

Most employees do not want to think about agent architecture. They want to give a task in natural language and get useful work back. A data analyst wants SQL queries, charts, and reports. A creative team wants mood boards, product photography variations, and campaign assets. A sales team wants account plans, customer context, follow-up materials, and next actions.

Role-specific plugins make Codex feel less like a blank AI tool and more like a specialized teammate.

For example, a data analysis plugin could let a user ask a question in plain English, connect to relevant data, generate SQL, run the query, build charts, and turn the result into an interactive report.

A creative production plugin could take a creative brief, generate a mood board, iterate product imagery, adapt versions for different scenarios, create display ads, and export assets into tools like Canva for final editing.

The key idea is that Codex is no longer only helping people write code. It is helping people produce work products.

2. Sites

The second major update is Sites.

Sites allow Codex to turn work into shareable, hosted, interactive websites or lightweight apps. This is one of the most interesting parts of the announcement because it changes what an AI output can be.

Instead of ending with a static document, a spreadsheet, or a slide deck, Codex can produce a living workspace.

OpenAI describes Sites as a way to turn ideas, analysis, and plans into:

• dashboards
• planners
• review workspaces
• project boards
• portfolios
• lightweight internal tools

These Sites can be shared by URL with people in the same workspace, giving teams a common place to review information, contribute feedback, track progress, and make decisions.

That matters because many business workflows do not fit neatly into one document or one spreadsheet tab.

Imagine asking Codex to create a Site for an upcoming customer review. It could generate an interactive page with product updates, open issues, usage trends, account context, and recommended next steps.

Or imagine asking it to build a scenario planner from a financial model. Instead of executives hunting through spreadsheet tabs, they could compare assumptions directly in an interactive interface.

Or a product marketing team could turn launch materials into a continuously updated work hub with messaging, milestones, owners, decision logs, and the latest approved assets.

The point is not that every team suddenly needs a website. The point is that AI-generated work can become more dynamic, collaborative, and operational.

3. Annotations

The third major update is annotations.

This may sound smaller than Sites, but it solves one of the most real problems in AI workflows: the first draft is easy; the revision process is hard.

Anyone who uses AI for work knows this pattern. The model creates something promising, but then you need to change one specific section, one chart label, one paragraph, one table, one design element, or one argument. If the only way to give feedback is through a general follow-up prompt, the model may over-edit or misunderstand what you want.

Annotations make feedback more precise.

Users can point to a specific part of the output and ask Codex to revise that area. For example:

• select a website navigation bar and ask Codex to change the font
• highlight a claim in an investment thesis and ask for the source or reasoning
• mark a chart in a slide and ask Codex to clarify the labels
• point to a paragraph in a document and ask for a sharper version
• select a table and ask Codex to reorganize it for executives

OpenAI has used similar interaction patterns for code, Markdown, and websites. Extending this idea into documents, spreadsheets, slides, and other work artifacts makes Codex feel much closer to a real collaborator.

The workflow becomes less like "generate once and hope it is right" and more like "review, comment, revise, and improve."

That is how teams actually work.

Why OpenAI is pushing this now

This move is not happening in isolation.

OpenAI is clearly accelerating toward the enterprise market. ChatGPT gave the company massive consumer distribution, but enterprise adoption requires something different: tools that fit into existing workflows, respect workspace controls, connect to approved systems, and produce useful business outputs.

Codex gives OpenAI a way to move from conversation to execution.

That is especially important because many companies are already past the early AI demo phase. They do not just want a model that can write a paragraph. They want AI that can help teams run processes, prepare materials, analyze data, generate internal tools, and keep work moving across departments.

The ChatGPT-Codex combination is OpenAI's answer to that demand.

ChatGPT provides the interface people already know. Codex provides the agentic execution layer. Plugins provide role-specific structure. Sites turn outputs into shared workspaces. Annotations make iteration more controlled.

Put together, this is a much stronger enterprise story than a general chatbot alone.

The Anthropic comparison

OpenAI is not the only company moving in this direction.

Anthropic has also been pushing enterprise agents, especially around high-value work in finance, engineering, design, and other professional functions. Its approach has often looked more vertical from the start: build agents for specific business scenarios, then expand from those workflows.

OpenAI's path is different.

OpenAI starts with the scale of ChatGPT, then pulls more execution capability into that familiar product surface. Instead of asking enterprise users to adopt a completely separate system, OpenAI can make the AI work agent appear inside a tool many employees already use.

That difference matters.

Anthropic looks enterprise-first and workflow-first. OpenAI looks distribution-first, then workflow-deep. Both strategies can work, but the competition will likely come down to the same question: which AI system can fit most naturally into the daily work of real teams?

What this means for users

For individual users, the shift means ChatGPT may become more useful for complex tasks that need output, not just answers.

Instead of asking for advice on how to build a dashboard, you may ask Codex through ChatGPT to build the dashboard. Instead of asking for a structure for a customer review, you may ask it to generate the review workspace. Instead of asking how to analyze sales data, you may ask it to connect to the data, run the analysis, and create the report.

For teams, the bigger impact is collaboration.

If Sites and annotations work well, AI outputs can become shared objects that teams inspect, revise, and use. That is more powerful than a private chat transcript. It turns AI work into something visible and actionable.

For businesses, the bigger question becomes governance.

If Codex can connect to tools, generate internal apps, access workplace context, and produce shareable outputs, companies will need clear rules around permissions, data access, review, security, and ownership. OpenAI's workspace controls and plugin approval flows will matter just as much as the model's raw capability.

The bigger picture

The ChatGPT-Codex integration points to a broader future for AI products.

The winner may not be the company with the smartest chatbot in isolation. The winner may be the company that turns AI into a dependable work layer across documents, data, design, code, communication, planning, and internal tools.

That is why this announcement feels bigger than a feature update.

OpenAI is trying to make ChatGPT the place where work begins and Codex the system that helps finish it.

If it succeeds, the way people use ChatGPT could change dramatically. It would no longer be just a place to ask questions. It would become a workspace where teams create, revise, deploy, and decide.

And that is where the AI agent race is really heading.

FAQ

Are ChatGPT and Codex becoming one product?

They are becoming more closely connected. ChatGPT remains the familiar conversation interface, while Codex is increasingly positioned as the agentic execution layer for building, analyzing, and iterating on work.

Is Codex still only for developers?

No. Codex is still strong for software engineering, but OpenAI is expanding it for broader knowledge work, including data analysis, creative production, sales, product design, investing, banking, reporting, and workflow automation.

What are Codex Sites?

Sites are shareable, hosted interactive webpages or lightweight apps created by Codex. Teams can use them as dashboards, planners, review spaces, project boards, work hubs, portfolios, or internal tools.

What are annotations in Codex?

Annotations let users point to a specific part of an output and ask Codex to revise that exact area. This makes feedback more precise than a broad follow-up prompt.

Why is this important for enterprises?

Enterprises need AI that can fit into real workflows, connect to approved tools, produce usable outputs, and support collaboration. The ChatGPT-Codex combination moves OpenAI closer to that kind of work platform.

Will this replace documents, spreadsheets, and slides?

Not completely. Instead, Codex may turn many documents, spreadsheets, and slides into more interactive, editable, and collaborative work artifacts.

What is the biggest risk?

The biggest risk is governance. As Codex becomes more capable across tools and workplace data, companies need strong controls for permissions, security, review processes, and data access.