Codex Sites Turns AI Work Into Hosted Workspace Apps
OpenAI's June 2 Codex update adds Sites, role-specific plugins, and annotations, moving Codex beyond software development into shareable internal apps and knowledge-work workflows.
OpenAI's latest Codex update is not just another coding-agent feature drop.
On June 2, 2026, OpenAI announced a broader Codex release aimed at making the system useful across roles, tools, and workflows. The concrete additions are role-specific plugins, Sites, and expanded annotations.1 The notable part is where this moves Codex: away from being only a software-development assistant and toward becoming a system for generating, refining, and sharing interactive work artifacts.
What changed
The headline number is that OpenAI says more than 5 million people use Codex every week.1 The company also says non-developers now account for about 20% of overall Codex users and are growing more than three times as fast as developers.1
That context explains the product direction. OpenAI introduced six role-specific plugins designed for different kinds of work: data analytics, creative production, sales, product design, public equity investing, and investment banking.1 According to OpenAI, those plugins bundle apps, skills, instructions, and workflows, and together include 62 popular apps and 110 skills.1
OpenAI also published a public GitHub repository for role-specific plugin templates. The repository describes the templates as packages of domain-specific skills, connector bindings, and starter assets that teams can customize for roles such as sales, data analytics, product design, and financial markets.2
That matters because it makes the workflow more concrete than a marketing phrase. These are not only prompts. They are packaged operating instructions and connector definitions that can be adapted to a team's own tools.
Sites are the bigger shift
The most interesting addition is Sites.
OpenAI says Sites are rolling out in preview for Business and Enterprise teams through the Codex app.1 In this preview, Codex can create and share interactive, hosted websites and apps that people in a workspace can access via URL.1
The examples OpenAI gives are not traditional software projects. They are dashboards, planners, review workspaces, project boards, galleries, lightweight tools, customer-review pages, financial scenario planners, and launch hubs.1 In other words, Sites target the messy layer between documents, spreadsheets, dashboards, and internal tools.
That is a practical area for AI because many teams already create one-off work artifacts that do not fit neatly into existing software. A forecast model may start in a spreadsheet, a launch plan may live across docs and slides, and a customer review may require data, next steps, owners, and open questions in one place. Codex Sites gives OpenAI a way to turn those mixed inputs into a hosted interactive surface.
OpenAI says Sites can be shared with anyone in the workspace and can be kept up to date as details change.1 The company also says it is working with early partners including Vercel, Wix, Base44, Replit, Lovable, Figma, Webflow, and Emergent as it builds toward a Sites partner ecosystem.1
Annotations close the loop
The third part of the release is expanded annotations. OpenAI describes annotations as a way to point Codex at a specific part of a result and ask for a targeted change.1
For software teams, that pattern is already familiar: select a section of a file, a component, or a website area and ask for a focused edit. OpenAI says the same interaction now extends to documents, spreadsheets, slides, charts, and Sites.1 That is important because generated work usually needs iteration. A first draft is useful only if the correction loop is precise enough to avoid starting over.
Why this is worth watching
The factual story is simple: Codex now has more users, more non-developer adoption, role-specific plugins, hosted interactive Sites, and annotation-based editing across more artifact types.
The significance is that OpenAI is treating Codex as a general work system, not only as a coding interface. Developers are still central, but the new release is aimed at analysts, marketers, designers, sales teams, investors, bankers, and operators too.1
That is a meaningful product shift. AI coding agents were first judged by whether they could write code. This release points to a broader test: whether an agent can gather context from business tools, generate a useful shared artifact, host it, and let a team refine it without turning every request into a software project.
For teams already living inside documents, spreadsheets, dashboards, and internal apps, that is a concrete new workflow to evaluate.