Codex is no longer only a tool for developers.
On 2 June 2026, OpenAI announced a new stage: Codex for every role, tool, and workflow.
Behind that very product-shaped title, the message is clear: OpenAI no longer wants to confine Codex to code. The company wants to turn it into a work agent usable by analysts, marketers, designers, sales people, finance teams, product teams and operations teams.
So the subject is no longer only:
“AI writes code”.
The subject is becoming:
AI moving directly into jobs, files, tools, dashboards, internal sites and company workflows.
And that shift matters far more than a simple tool update.
An agent already used by 5 million people every week
OpenAI states that more than 5 million people now use Codex every week.
Another important figure: non-developers already account for roughly 20% of Codex users.
And their growth is said to be more than 3 times faster than that of developers.
That is probably the most revealing number in the announcement.
Codex started out as a software development agent.
But OpenAI is now seeing something else appear: users who do not necessarily write code, but who want to delegate work.
Build a dashboard.
Prepare an analysis.
Turn a brief into a deliverable.
Build an internal page.
Summarise documents.
Generate materials.
Connect tools.
Prepare a report.
The code becomes almost secondary.
What counts is that Codex can produce something usable.
This is the same movement, seen from the other end, as the one we described when developers started becoming supervisors of AI.
What actually changes
OpenAI is announcing three main additions:
- plugins by job function;
- Sites shareable inside the company;
- annotations to correct deliverables precisely.
This is not a simple convenience improvement.
It is an attempt to move Codex out of the developers-only world.
Plugins by job function
OpenAI is launching six specialised plugins.
Together they cover 62 popular applications and 110 skills.
The areas OpenAI cites include:
- data analysis;
- creative production;
- sales;
- product design;
- equity investment;
- investment banking.
The idea is simple: instead of asking Codex to work everything out from a prompt, you give it blocks of skills fitted to a job.
For an analyst, Codex can help turn data into a summary or a dashboard.
For a sales team, it can prepare client accounts, organise information, or produce materials.
For a marketing or creative team, it can go from a brief to several variants of a deliverable.
For a finance team, it can structure analyses, compare companies or prepare documents.
The important point is this:
Codex is no longer only an agent that edits code. It is becoming an agent that operates business tools.
”Sites”: Codex builds interactive spaces
Another addition: Sites.
OpenAI presents this feature as a way to create interactive spaces for Business and Enterprise teams.
Codex can produce:
- small internal sites;
- simple applications;
- dashboards;
- planners;
- project boards;
- review spaces;
- deliverables shareable by URL inside a workspace.
The feature is in preview.
But it shows the direction clearly.
Instead of asking an AI to generate an isolated text or file, the user can ask it to create a complete workspace.
A dashboard.
A tracking interface.
A review page.
A small internal tool.
That is exactly the kind of task which, until now, often required a developer, a no-code tool, an advanced spreadsheet or several rounds between teams.
Annotations: correcting one precise part of the work
OpenAI is also adding a system of annotations.
The principle: the user can point at a precise part of a deliverable and ask for a targeted change.
For example:
- fix a sentence;
- change a chart;
- adjust a navigation bar;
- change a section of a site;
- revise part of a document;
- rework a slide;
- change a cell or a piece of logic in a spreadsheet.
This matters, because real professional work is not just producing a first version.
Real work is usually:
- produce;
- review;
- correct;
- adjust;
- redo part of it;
- approve.
With annotations, OpenAI is trying to bring Codex closer to a genuine work cycle.
Not just a generation.
An iteration.
Codex can also act inside Windows
On 29 May 2026, OpenAI also announced an important update: Codex supports Computer Use on Windows in the Codex app.
In practice, eligible users can ask Codex to:
- see a Windows application;
- click;
- type;
- test;
- debug;
- refine whatever is being built.
OpenAI also says the user can start a job on a Windows machine, then follow or steer its progress from ChatGPT on iOS or Android, or from Codex on Mac.
The Windows machine remains the host of the files, the terminal, the local server and the project context.
An important point: at launch, this feature is not available in the European Economic Area, the United Kingdom or Switzerland.
That regional limit should be taken seriously.
So not everything is available everywhere.
The real change: the agent moves into workflows
What this announcement shows is not only a new feature.
It is a strategy.
OpenAI wants to make Codex a layer of work that cuts across tools.
Before, Codex was mainly there to:
- read code;
- edit a repository;
- fix bugs;
- open a pull request;
- help developers.
Now, OpenAI is pushing Codex towards:
- documents;
- spreadsheets;
- dashboards;
- internal sites;
- business tools;
- company workflows.
That drift is major.
Because it brings AI close to a far wider area than software development: the daily work of teams.
It is the same direction we followed when AI tools stopped answering and started acting — and what each of these agents actually holds up to on long tasks is what our AI agent ranking tracks.
What this can change for companies
The obvious benefit is productivity.
An agent able to prepare a report, build a dashboard, structure an analysis or produce an internal site can save a lot of time.
But there is also a risk.
When an AI writes a false paragraph, it is a mistake.
When it updates a dashboard, prepares a pitch, transforms a brief, changes a workflow or creates an internal tool, the mistake can become operational.
It can end up in a decision.
In a presentation.
In an analysis.
In a report.
In a process.
That is where the subject becomes more serious.
Codex is not only leaving the code.
It is starting to enter the production chain of work itself.
No need to overstate it
Codex is not replacing employees yet.
And OpenAI is not saying that every job can delegate anything without checking.
Several features are still in preview.
Some are reserved for Business or Enterprise customers.
Some are limited by region.
And as always with AI agents, the result has to be verified.
So the real subject is not:
“Codex is going to replace everyone.”
The real subject is more precise:
Codex is starting to turn office tasks into tasks that can be delegated to an agent.
That alone is enough to deserve attention.
What to take away
Codex was born as a coding agent.
In 2025, OpenAI presented it as an agent able to work in a cloud environment, write code, fix bugs and open pull requests.
In June 2026, the wording changes.
OpenAI now talks about roles, tools, workflows, business plugins, interactive sites and annotations on deliverables.
The signal is clear:
the AI agent is gradually leaving the code editor to move into the daily work of teams.
Today, it still assists.
Tomorrow, it could produce a much larger share of the work.
And that is exactly what makes this announcement important.
Not because Codex can do everything.
But because OpenAI is starting to place it everywhere work gets made.