OpenAI Puts Codex in ChatGPT With 6 Business Plugins

OpenAI Codex plugins moved the agent out of the terminal and onto the desk of every business user. On June 2, 2026, at an "Intelligence at Work" livestream, OpenAI announced Codex for every role — six role-specific plugins, a preview of Codex Sites, and an Annotations feature — and confirmed Codex is coming to the ChatGPT app itself. The story is not a bigger toolbox for developers. It is the moment agentic work became a native skill for analysts, marketers, and ops teams.

The numbers OpenAI shared frame why this matters. Codex has passed more than 5 million weekly active users, and the company reports that knowledge workers — analysts, marketers, designers, researchers, investors, and bankers — now make up roughly 20% of that base and are growing more than three times faster than the developer population. Those figures are OpenAI's own telemetry, not independently audited, but the direction is unmistakable: the fastest-growing Codex user is no longer a software engineer.

This guide covers what actually shipped on June 2 versus what is still on the roadmap, a coverage map of which role dependency each plugin removes, the real enterprise limitation in Codex Sites that launch-day coverage skipped, and a concrete playbook for agencies and operations teams. Where a number comes only from OpenAI's usage report, we label it as such — and where the in-product integration has been announced but not yet delivered, we say so plainly.

The June 2 announcement bundled four things at once: the six role-specific plugins, Codex Sites in preview, the Annotations feature for surgical edits, and the stated intent to put Codex directly inside the ChatGPT app. OpenAI framed the rationale simply — many business users already know ChatGPT but have never discovered Codex, so meeting them where they already work is the distribution play. The in-ChatGPT integration is the part to watch your expectations on: it was described as arriving "in the next few weeks," with no committed date.

Underneath the plugins is a model OpenAI calls GPT-5.3-Codex, which it describes as its most capable agentic coding model to date, combining frontier coding performance with broader reasoning and professional knowledge and running roughly 25% faster than the prior generation. The capability story alone is not new — what changed on June 2 is the packaging. Instead of a coding assistant for engineers, OpenAI is shipping pre-built professional roles a non-technical user can invoke from a chat box.

Each plugin bundles a set of integrations, role-specific instructions, and contextual information designed to simulate a particular professional. OpenAI says the six plugins together aggregate 62 popular business applications and 110 automated skills out of the box. The launch set leans toward analytics, creative, sales, design, and finance — and five more roles (Corporate Finance, Private Equity Investing, Marketing Strategy, Strategy Consulting, and Legal) are on the roadmap but not yet released.

The remaining launch plugins extend the same pattern into design and finance. Product Design targets the prototype-to-spec loop; Public Equity Investingdraws on Moody's, FactSet, and S&P data for earnings review, company comparisons, and tracking an investment thesis; and Investment Banking rounds out the launch set. The shape across all six is consistent: each one collapses a workflow that previously needed a specialist plus their tooling into a single natural-language session.

Most coverage lists the plugin names. The more useful question for a CMO or ops director is: which person on my team now has fewer dependencies? The table below maps each launched plugin to the role it simulates, its headline integrations, and the handoff it removes. Integrations are drawn from the launch coverage; the "handoff removed" column is our reading of the workflow implication, not an OpenAI claim.

Codex Sites, in preview for Business and Enterprise ChatGPT plans, turns Codex output into hosted, interactive apps — dashboards, planners, review workspaces, project boards, and galleries — shareable within a workspace via URL at no additional cost. OpenAI names Wix, Base44, Replit, Lovable, Figma, Webflow, and Emergent as ecosystem partners. On paper it is the feature that turns an analysis into a tool your whole team can click into.

The catch is authentication. Sites uses "Sign in with ChatGPT," and as observed at launch it does not support SAML, custom identity providers, or external partner access — viewers must be members of the ChatGPT workspace. For an agency that bills clients by deliverable, that is a hard constraint: a client who needs to see a Codex-built dashboard would have to be inside your workspace, or you fall back to screenshots and exports. This is a genuine limitation versus standalone no-code builders, and it was largely absent from the launch-day enthusiasm.

The Annotations feature is the quieter half of the canvas story. It lets a user select a specific section of a document, spreadsheet, slide, or image inside Codex and give a targeted edit instruction — a surgical rewrite rather than a full regeneration. For anyone who has watched an AI tool rebuild an entire artifact to fix one line, this is the practical detail that makes Codex usable for real iterative work.

OpenAI's "Next Era of Knowledge Work" report is the source for the adoption story, and every figure below is its own telemetry — useful for direction, not independently verified. The headline most outlets ran with is absolute growth: 5 million-plus weekly users, up more than six times since the desktop app launched in February 2026. The more interesting read is composition. By OpenAI's account, knowledge workers are growing more than three times faster than developers, and personal users more than four times faster — which means the developer share of the base is being diluted, not just out-grown.

Here is the part the coverage under-states. The dominant framing is "Codex is expanding beyond developers." The sharper read is the inverse: the barrier to agentic work just collapsed for every knowledge worker. A designer can ship a prototype without a developer. An analyst can stand up a dashboard without filing an engineering ticket. A sales lead can prep an account and update the CRM without routing through ops. In each case the work still happens — but the "send it to a specialist and wait" step does not.

That removed step is a role, not just a task. For years the implicit org chart of digital work ran analyst to developer to deployment, with a technical liaison in the middle translating intent into something a specialist could execute. The plugin model dissolves that middle. The implication agencies should sit with is uncomfortable and useful at the same time: the value of being the translator drops, and the value of designing the workflow around the new capability rises.

Projecting forward, the constraint moves from access to judgment. When anyone on a team can generate a dashboard, a campaign board, or a contract draft, the differentiator stops being who can operate the tool and becomes who knows which artifact is worth making, which output is wrong, and which process should be redesigned around the agent. OpenAI's own report invokes the classic productivity paradox — the observation that technology delivers its largest gains only after organizations redesign their processes around it. The plugins lowered the tool cost. The gain is locked behind the redesign, and that is squarely what agentic-first delivery actually means for agencies.

On the same day, DocuSign announced a ChatGPT and Codex app that puts its Intelligent Agreement Management workflows behind a natural language prompt, available globally in English. Almost every outlet led with the six plugins and treated this as a footnote. It is the opposite of a footnote. The legally binding agreement loop — generate a contract, route it for signature, archive it — becomes something you ask for in plain language. For an ops team that runs contracts, that removes the "send to legal, wait three days" step from the middle of a deal.

DocuSign is not a small partner to onboard this way. By its own account it serves more than 1.8 million customers across 180-plus countries, with fiscal 2026 revenue of $3.2 billion — a full fiscal-year figure, not a run-rate, since its fiscal year ends in late January. The app spans Legal, Sales, Procurement, HR, and Finance agreement workflows. Pair it with the Sales plugin's CRM reach, and you can see the assembled loop: find the account, prep the deal, generate the agreement, and route it for signature without leaving the chat surface.

The strategic context behind all of this is OpenAI's push to land inside enterprises directly. In May 2026 it launched a dedicated deployment company with more than $4 billion in initial investment from a roster of global investment firms, consultancies, and system integrators, and folded in an applied-AI consulting team. The June 2 plugins are the product surface of that strategy: not just better models, but pre-packaged roles and partner integrations aimed at getting agentic work into the daily workflow of non-technical staff.

The announcement does not require an immediate platform switch — much of it is still rolling out — but it does change the planning calculus. The four moves below sort the response by where the value and the risk actually sit.

For teams already running agentic pilots, the practical next step is sequencing rather than a single decision. Slot the plugin evaluation into a staged rollout — pilot, measure, then expand — rather than a big-bang adoption; that is exactly how to build agentic workflow adoption in 30-60-90 days. If your function is marketing specifically, the role-by-role view matters more than the platform: see the marketing team AI automation playbook and a role-by-role breakdown of AI tools for marketing teams. And when the question becomes whether to adopt OpenAI's pre-built plugins or build your own agent, work through the enterprise build-vs-buy decision for AI agents before committing budget. If you want a partner to run that evaluation with you, our AI transformation engagements and CRM automation work start with exactly this kind of workflow mapping.