OpenAI is putting Codex in every ChatGPT app, with six business plugins for non-coders

OpenAI wants Codex in front of people who can’t write code. On June 2 the company said it’ll fold Codex into the ChatGPT app everywhere within the next few weeks, and shipped six role-specific plugins built for sales reps, analysts, designers, and finance teams rather than engineers. The pitch is plain: stop treating Codex as a developer tool.

That’s a sharp turn from where Codex was three weeks ago. In mid-May, OpenAI’s big Codex move was putting the agent inside the ChatGPT mobile app so a developer could approve a diff from a phone while the agent ran on their laptop. That was still a coder’s feature, just on a smaller screen. This launch is the next step, and it points in a different direction entirely. The audience OpenAI is now chasing isn’t developers at all. It’s the hundreds of millions of people who open ChatGPT every week and have never once opened Codex.

What OpenAI actually shipped

Three things landed on June 2, and they fit together. First, the six plugins: data analytics, creative production, sales, product design, equity investing, and investment banking. Each one wires Codex into the apps and skills a given job already lives in, so the agent shows up understanding the tools of that role instead of staring at a blank repo. OpenAI says the six together bundle 62 popular apps and 110 skills.

Second, annotations. You highlight a specific chunk of a document or file and tell Codex what to do with that exact part, rather than describing the passage in a prompt and hoping it picks the right one. As OpenAI put it, “Developers already use annotations in Codex to refine code, Markdown files, and websites Codex creates.” The June change extends that editing model to the new business surfaces.

Third, Sites. Codex can now publish its output as a hosted interactive website with a shareable URL, instead of handing you a local file you have to deploy yourself. OpenAI lists Wix, Base44, Replit, Lovable, Figma, and Emergent as launch partners. The practical effect: a sales lead can ask Codex to build a deal dashboard and send a link, no hosting step, no developer in the loop. OpenAI’s own framing is that Codex can take “ideas, analysis, and plans” and turn them into dashboards, planners, review workspaces, project boards, galleries, and lightweight tools. In other words, the output isn’t code anymore. It’s a thing a non-technical colleague can click.

Put the three together and the shape is clear. The plugins decide what Codex knows about a role, annotations control how you correct it, and Sites determine what it hands back. Each piece on its own is incremental. Bundled, they describe a Codex that a non-engineer could actually use start to finish, from prompt to shipped artifact, without ever seeing a line of code.

All three roll out in preview to Business and Enterprise customers first. OpenAI hasn’t committed to a consumer date.

Why aim a coding agent at non-coders

Because the growth math says to. Codex passed 5 million weekly active users, up more than 6x since the desktop app launched in February. That’s a strong curve. But it’s a rounding error against ChatGPT’s overall base, and OpenAI knows exactly where the untapped users are: inside the app they already pay for.

Folding Codex into the main ChatGPT app removes the single biggest barrier, which was that you had to know Codex existed and go install or pair it. A marketing manager isn’t going to download a separate coding tool. She will, however, notice a new capability inside the app she opens 10 times a day.

There’s a strategy bet underneath the product one. OpenAI’s chief revenue officer, Denise Dresser, framed it this way: “AI is becoming capable of doing increasingly meaningful work inside organizations. The challenge now is helping companies integrate these systems into the infrastructure and workflows that power their businesses.” Read that as the company conceding that raw model capability isn’t the bottleneck anymore. Distribution and workflow fit are. The plugins are the workflow-fit play.

The pattern isn’t unique to OpenAI. Enterprises have been wrestling with how to put agentic coding tools in front of large teams without the bill exploding. Uber’s engineering org found that out the hard way when its AI tooling budget blew past plan on Claude Code usage. Microsoft went the other way and pulled Claude Code licenses from teams it judged weren’t getting the value. OpenAI’s answer to that same tension is to make the agent useful to far more than just engineers, so the per-seat economics work across a whole company, not one department.

What “for every role” really means

The phrase is doing a lot of work, so look at the roster. The six shipped plugins cover sales, data, creative, product design, and two flavors of finance. OpenAI says the next wave includes Corporate Finance, Private Equity Investing, Marketing Strategy, Strategy Consulting, and Legal. Stack those up and you’ve got a map of high-salary knowledge work, the exact roles that cost companies the most and that OpenAI would most like to sell agent seats into.

That’s the honest read on this launch. It isn’t a coding update. It’s OpenAI trying to turn Codex from a developer product into a company-wide one, the way ChatGPT itself made the jump from a curiosity to a default work tool. Whether a “Sales” plugin actually does sales work, or just produces plausible-looking dashboards a human still has to check, is the question the preview will answer.

The 62-app, 110-skill count OpenAI put on the plugins is worth a second look too. It signals that these aren’t thin wrappers around a system prompt. Each plugin connects to real tools a role uses, which is the hard part of making an agent useful at work. A data-analytics plugin that can’t reach the warehouse is a toy. The harder it integrates, though, the more access it needs, and the more a bad output can touch. OpenAI is betting that companies will trade that risk for the productivity, the same trade the whole agentic-coding wave has been asking developers to make for a year.

One caveat worth keeping front of mind: agents that act inside business systems raise the stakes on mistakes. A wrong diff in a code review gets caught by tests. A wrong number in an investment-banking deck, published straight to a shareable Site, has fewer guardrails. OpenAI hasn’t said much yet about review controls for the non-developer surfaces, and that gap matters more the further Codex gets from engineers who know to distrust the output.

What this means for you

If you’re a developer, not much changes today. Your Codex workflow is the same one you had after the mobile launch in May. If you run or buy tools for a business team, this is the launch to watch. Get on the Business or Enterprise preview, hand a real task to the plugin closest to your team’s work, and grade the output against what a competent human would produce, not against a blank page. The plugins will demo well. The thing to measure is whether the artifact survives a second look. And watch the next wave: when the Legal and Strategy Consulting plugins ship, you’ll know whether OpenAI is genuinely expanding what Codex can do or just renaming the same agent for a longer list of buyers.