The Inversion: Why OpenAI Is Letting Non-Developers Build What IT Used To

OpenAI has a problem it will not name aloud. Its coding tool, Codex, is winning the wrong race.

OpenAI now has more than 5 million weekly active users, up from 4 million in April and 3 million before that. But the more interesting number is the one OpenAI buried in the same announcement: roughly 20 percent of those users are not developers. They are analysts, marketers, bankers, salespeople, designers. And that slice is growing more than three times as fast as the developer base.

That is the actual story. Not that Codex exists, or that it writes code well, but that it has become something OpenAI did not plan for and does not fully control: a tool that non-programmers want, for work that has nothing to do with programming.

OpenAI shipped six new role-specific plugins — bundles that connect Codex to the apps knowledge workers already use. The data analytics plugin hooks into Snowflake, Databricks, Tableau. The sales plugin surfaces Salesforce and HubSpot. The investment banking plugin touches FactSet, PitchBook, LSEG. Sixty-two apps in all, one hundred and ten skills, all configured to work with Codex out of the box. No code required.

Historically, enterprise software worked the other way around. A company bought a platform — Salesforce, SAP, Workday — and its employees spent months learning the tool's logic, its menus, its quirks. The human adapted to the software. OpenAI is now building the reverse: software that configures itself around the human, around the role, around the specific workflow a person already has. That is a genuine inversion, and it is not small.

The tension is that OpenAI is chasing this market from a position of weakness in the one it originally owned. Among developers, Anthropics Claude Code is still preferred. Ars Technica put it plainly in March: talk to developers and you find more Claude Code users than Codex users. OpenAI is playing catch-up on the plugin ecosystem Claude Code shipped first. This is not a secret inside the industry.

So OpenAI is doing what large software companies do when they cannot win on product alone: it is buying distribution. The company brought in Accenture, Capgemini, and PwC as consulting partners to sell Codex to enterprise clients. Denise Dresser, OpenAIs chief revenue officer, said the partners will help companies bridge that gap between how to use it, how to expand it, and how to move even more quickly. She has a personal Codex agent called Chief that summarizes meetings, writes notes, and updates Salesforce and Slack. She is using the product. That is fine. It is also not the same as a Fortune 500 signing a contract.

The inversion OpenAI is describing — software that learns the shape of professional work rather than demanding professionals learn the shape of software — has happened before. Spreadsheets did not start in the finance department. The first spreadsheet users were scientists and engineers. By the mid-1980s, every department in every company had someone who knew Excel better than their job description required, because the tool had reorganized how work got done. CAD tools spread from aerospace to architecture to industrial design for the same reason: once a tool becomes expressive enough, people find uses its creators did not anticipate.

The difference, and it is a real one, is velocity. The spreadsheet analogy unfolded over a decade. Codex is at 5 million weekly users in under two years. The plugin ecosystem OpenAI announced Monday is a deliberate acceleration of that diffusion — a bet that the next million Codex users will not be developers at all, but the analyst who builds her own dashboards, the banker who automates her own comps, the product manager who prototypes her own designs.

OpenAI will not say how many of the new plugins are live versus preview. The knowledge worker growth figures come from an OpenAI research report the company has not made freely available. The enterprise consulting deals are real — Accenture, PwC, and Capgemini do not work for free — but whether they have produced actual contracts or only pipeline is not public. These are the gaps the wire left out, and they are the reason to be careful about declaring victory before the deals close.

What is clear is the direction. OpenAI built Codex to replace programmers. The programmers did not fully embrace it. The people who embraced it were everyone else. Now OpenAI is building toward that world deliberately, with plugins, partnerships, and a product strategy that sidesteps the dev community it originally targeted. Whether it can close the enterprise deals fast enough to matter before Anthropic or Google does the same thing in a different color is the next question. That one is not answered yet.