Codex goes beyond code: 5M weekly users in June, 40M by October

Codex is no longer just a tool for programmers. In June, the weekly audience of the OpenAI product passed 5 million people, and knowledge workers grew fastest: analysts, managers and operations teams. Their growth outpaced developers by a factor of three. By October, Codex and ChatGPT Work together had 40 million active users. The Chinese outlet Jiqizhixin ties this trend to a topic the industry is now debating: Pro-class agents that can take on entire workflows.

The audience has shifted from programmers to the office

Codex started out as a coding agent: it reads a repository, edits files, runs tests and proposes a pull request. That mechanic transfers easily to any work with digital artifacts. A spreadsheet, a presentation, a report and a CRM export all work the same way: there is a set of files, a goal and a verifiable result.

That is why knowledge workers came to the product faster than anyone expected. They don't need to write programs. They only need to set a task, grant access to work data and check the outcome. Growth three times faster than among developers shows that demand for agents in office professions had been building for a long time, and has now simply found a suitable interface.

What is behind the 40 million figure

The 5 million in June and the 40 million in October cannot be compared directly. The first figure covers only Codex; the second combines Codex and ChatGPT Work. These are different products with overlapping audiences, and some users almost certainly use both. The real growth of Codex itself cannot be derived from this data.

The strategy, however, is visible. OpenAI is assembling a coding agent and a work assistant into a single product line, where the boundary between a "programmer's helper" and a "digital employee" is gradually blurring. For the company, this is a way to sell subscriptions and enterprise licenses not only to technical departments but to everyone else.

Key metrics from the article:

• June 2026: more than 5 million weekly active Codex users;
• knowledge-worker audience growing 3 times faster than developers;
• October 2026: 40 million active users of Codex and ChatGPT Work combined.

Pro Agent and the Chinese context

Against this backdrop, Jiqizhixin writes about the appearance of the "industry's first Pro Agent" and about Kevin Knight's visit to China. He is called the "progenitor of AI" in computational linguistics and is known for his work on statistical machine translation. The topic of his talk is the next leap in AI development. The headline itself shows where the conversation is shifting: from chatbots to agents that work autonomously and for longer.

For the Chinese market, this is a signal that competition is now over the "professional agent" category. The winner is whoever establishes itself in everyday workflows before the others, not whoever scores higher on a benchmark.

Where to start for teams that count employee time

The practical takeaway is simple: agent pilots should be launched beyond engineering. Good candidates are tasks with a clear structure and a verifiable result: weekly reports, data reconciliation, drafting documents, sorting through customer requests.

Before launch, teams should agree on three things. First: what data the agent can access and where the boundaries lie. Second: who checks the output and against what checklist. Third: how to measure the effect, for example in hours saved per task rather than the number of model requests. Companies that settle this before others will get real returns from agents, and the rest will end up with expensive subscriptions that people use once a month.