OpenAI’s Codex Is Now Your Sales Team, Data Analyst, and Creative Director

Executive Summary

• Fact: On June 2, OpenAI launched six role-specific plugins for Codex — bundling 62 apps and 110 skills across sales, data analytics, creative production, product design, equity investing, and investment banking.

• Impact: Knowledge workers already represent 20% of Codex’s 5 million weekly active users and are growing three times faster than the developer base — a deliberate platform pivot from coding assistant to general-purpose business tool.

• Surprise: The Creative Production plugin can turn a marketing brief into ecommerce-ready image sets and display ad variations using Figma, Canva, and Shutterstock — work that currently lives inside agency retainers or internal design teams.

In February, OpenAI launched a desktop app for Codex. Four months later, 5 million people use it every week — and one in five of them isn’t a developer.

That’s the number OpenAI chose to lead with on June 2. Knowledge workers now make up 20% of the user base, growing three times faster than the engineering cohort that built Codex’s early reputation. The pivot is no longer implicit in usage data. It’s company policy. Codex isn’t a coding tool anymore. It’s a work platform.

Six Plugins, One Strategic Bet — and the Creative One Is the Wildcard

The center of Tuesday’s announcement is six role-specific plugins, each bundling its own stack of integrations, instruction sets, and contextual permissions. Combined, they connect to 62 apps and 110 discrete skills.

The data analytics plugin hooks into Snowflake, Databricks Genie, Hex, and Tableau. Ask it why a key metric dropped last week, and it builds the analysis. The sales plugin connects to Salesforce, HubSpot, Outreach, Clay, and Slack — it can identify high-priority accounts, draft follow-up messages, update CRM records, and flag deals at risk, all within a single Codex session.

Then there’s the creative production plugin, which is where brand and marketing teams should pay the closest attention. Connect it to Figma, Canva, Shutterstock, Picsart, and Fal; hand it a brief; and it generates campaign boards, display ad variations, and — the part that stops e-commerce operators mid-sentence — ecommerce-ready product image sets. Lifestyle shots. Formatted for platform specs. Without a photographer on retainer.

Two finance-focused plugins complete the six: one for public equity investors (connecting to Moody’s, FactSet, PitchBook, LSEG, and S&P), one for investment bankers preparing pitch materials and turning diligence into client-ready recommendations. OpenAI has already signalled that Finance and Legal verticals are next on the roadmap.

What’s striking about this architecture is how deliberately OpenAI has avoided the classic enterprise AI trap: one model, every use case, middling results for all of them. By giving each plugin its own approved tool stack, its own instruction layer, and its own permissions model, OpenAI is implicitly admitting that context is the product now — not just raw model capability.

5 Million Users and the Signal Inside the Number

Internal data from OpenAI’s June 2 report, “The Next Era of Knowledge Work,” shows data analysis use among knowledge workers up 110% since the February desktop launch. The fastest-growing tasks across non-developer users are data analysis, research, and what the company calls “knowledge artifact creation” — reports, spreadsheets, presentations, and contracts that previously required either skilled humans or a string of disconnected tools.

This is a significant data point for any COO or brand leader watching the agentic commerce frameworks emerging across retail. The 110% growth in data analytics use doesn’t reflect a niche use case. It reflects a workflow that most marketing and operations teams run every week, and that most of them still run manually.

Thinking about deploying AI agents across your business teams? See how Transform by Epinium redesigns workflows around AI — not just adds tools on top →

What the Sites Feature Actually Changes About How AI Work Gets Delivered

Alongside the plugins, OpenAI introduced a feature called Sites. Codex can now publish its output as a hosted, shareable interactive webpage — not a local file, not a chat response. Launch partners include Wix, Base44, Replit, Lovable, Figma, and Emergent.

Easy to overlook. Operationally significant. A sales analyst asks Codex to build a competitive comparison. Instead of a spreadsheet sitting in someone’s Downloads folder by end of day, Codex produces a live URL the whole team can access. A marketing manager requests a campaign overview with asset previews. The output isn’t a document. It’s a page. Combine that with the new Annotations feature — which lets users designate specific sections of a file to give Codex targeted instructions — and what OpenAI is building starts to resemble a lightweight project management layer with AI execution at its core.

There is, however, a contrarian read worth sitting with. Last week, Uber disclosed that it had capped employee AI spending after exhausting its annual budget in just four months. Uber had actively encouraged maximum AI use; the result was uncontrolled cost growth before any governance structure was in place. OpenAI’s plugins solve for capability. They don’t solve for governance. A COO who deploys Codex enterprise-wide and finds three months later that six teams have independently built overlapping intelligence workflows — with no shared data model and no audit trail — will recognize the Uber problem wearing a different face.

Epinium data

In onboarding audits conducted across more than 400 brand accounts on Epinium Platform over five years, content creation and data analysis tasks consume an average of 58% of marketing team hours in the first 90 days. The workflows OpenAI’s new plugins now target are not peripheral to brand operations — they are the operational core, and the brands that restructure their teams around them first will compound the advantage over time.

If You’re Running Teams, Not Just Using Tools

The questions worth sitting with aren’t about whether to evaluate the Codex plugins. They’re sharper than that.

Which roles in your organization are being materially reframed by this release — not replaced outright, but redefined? Your data analytics function is an obvious candidate. Your creative operations team is another. If the Sales plugin handles CRM updates, follow-up drafting, and deal risk flagging, what does that imply for rev-ops headcount decisions over the next 18 months?

What we’re seeing at Epinium is that the brands moving fastest aren’t trying to automate everything simultaneously. They pick one high-friction workflow, run a structured pilot with measurable baselines, and expand from there. The firms that attempted company-wide AI deployment in Q1 and are now quietly cleaning up share a consistent pattern: they skipped the governance layer entirely.

OpenAI is giving your teams remarkable new capabilities. The competitive advantage isn’t in having the tools — everyone will have them. It’s in knowing how to use them with discipline, and knowing when stopping is the smarter call.

Are the Codex plugins available to all users, or only enterprise tiers?

As of June 2, the six plugins are rolling out broadly to Codex users, though full access to all partner integrations — including Salesforce, Snowflake, and FactSet data connections — may require existing enterprise-tier agreements with those platforms. OpenAI has not published a standalone pricing tier for plugin access. Check your current ChatGPT or Codex subscription for availability, and contact your account representative for enterprise rollout timelines.

Do you need a developer or IT team to deploy the Creative Production or Sales plugins?

No — each plugin bundles its own integration logic by design. Connecting to Salesforce or HubSpot requires standard OAuth authorization, not custom development work. That said, any organization with no prior API governance should plan for some friction when authorizing data access across six or more platforms simultaneously. A lightweight IT or operations review of what Codex can access — before deployment — is well worth the hour it takes.

Can the Creative Production plugin handle multilingual e-commerce assets at scale?

OpenAI’s announcement materials don’t address this directly. The plugin’s connected tools — Canva, Figma, Picsart — all support multilingual asset creation natively. Whether Codex reliably orchestrates localized campaign briefs across, say, five language versions simultaneously depends on how the prompt layer handles language switching — which remains an open question. Test this specific workflow carefully before committing a multilingual production pipeline to Codex.

When does AI-assisted work become AI-displaced work, and how do you manage the transition?

A useful decision frame: if a task takes a skilled person under two hours, is fully repeatable, and produces a verifiable output, AI can handle it today. If it requires judgment across ambiguous signals, stakeholder relationships, or institutional context that exists nowhere in a document, a human remains significantly better. The Codex plugins are in the first category for specific sub-tasks within each role. The productive response is to redefine roles around the judgment layer — the interpretation, the decision, the client relationship — rather than the execution layer those roles used to own.

How does Codex compare to Microsoft Copilot for teams already in the Microsoft 365 ecosystem?

Copilot’s primary advantage is depth of integration within the Microsoft stack — Word, Excel, Teams, Outlook — with no new application to adopt. Codex’s advantage is the specialist plugin model: a sales team gets a tool purpose-built for sales workflows with its own curated integrations, rather than a general assistant layered over existing spreadsheets. For companies already paying for Copilot licenses, the honest answer is to run both in parallel for 90 days on equivalent tasks and measure actual output quality. The winner will depend on your workflow, not on brand loyalty.

OpenAI’s pivot to knowledge workers is not a signal that developers matter less. It’s a signal that the total addressable market for AI tools just expanded by an order of magnitude — and that every business function, from your marketing ops manager to your head of sales, is now inside the blast radius.

Ready to bring AI execution into your business teams with the right governance in place? Transform by Epinium embeds dedicated AI directors inside your organization to map workflows, identify high-value automation opportunities, and build the oversight layer that platforms like Codex require to deliver on their promise. Book your free 30-minute AI diagnosis →