OpenAI built Codex as a tool for software engineers. Today it is openly positioning the product as something closer to a general-purpose work assistant for the office. The signal is clear in the numbers: Codex now has 5 million weekly users, up more than six times since a desktop app launched in February. Non-technical workers represent about 20 percent of that base and are growing more than three times as fast as the developer core.
The six new job-specific tools cover data analytics, creative production, sales, product design, equity investing, and investment banking. Each one bundles relevant instructions and connections so a person in that role can get useful output without setting anything up from scratch. The tools are designed to work out of the box, while also improving as users customize them over time.
Two other additions round out the update. A new Sites feature lets Codex turn its outputs into a live, interactive webpage rather than a static file, with partners including Figma, Wix, and Replit already connected. An Annotations feature lets users point directly at a part of a document, giving Codex more precise instructions about exactly what to act on.
What is the actual job these tools are replacing? Research from McKinsey estimates the average office worker spends roughly 28 percent of their working week on email and close to 20 percent just searching for information buried across different systems. The fastest-growing uses inside Codex right now are data analysis, up 110 percent week over week; market research, up 37 percent; and the creation of reports and memos. About half of all Codex users are now running more than one task at the same time, up from under a third in April.
The competitive context matters here. Anthropic, OpenAI's closest competitor, moved into enterprise territory earlier and more systematically. It launched dedicated agents for financial services professionals in May, covering tasks like building pitch materials, screening compliance files, and closing monthly accounts. It had previously done the same for healthcare and legal. Anthropic has had a head start on signing up banks including Goldman Sachs, Citadel, and Citi.
OpenAI is catching up through a different structure. The Deployment Company, a $4 billion joint venture launched three weeks ago, is designed to send teams of engineers directly inside large businesses to install and run OpenAI tools in their operations. The 19 investors in the venture, including TPG, Brookfield, Bain Capital, and McKinsey, also happen to control more than 2,000 businesses between them, providing a built-in pool of potential clients. Priority sectors are healthcare, logistics, manufacturing, and financial services.
For anyone running a business, the practical question is whether these tools are genuinely useful now or still experimental. The honest answer is: it depends heavily on how clean your data and internal processes already are. Tools that analyze spreadsheets or draft research reports work well when the underlying information is structured and reliable. When it is not, errors compound quickly. A sensible approach is to identify one or two high-repetition, low-stakes tasks, run Codex on them for a few weeks, and measure whether the output quality justifies the time saved on review. The riskier move is to plug these tools directly into customer-facing or regulated work without a human checkpoint in the process.
The broader shift is real regardless of which product wins. Office work built around moving information, writing documents, and assembling analysis is the part of professional life most exposed to automation right now. The companies that figure out how to use these tools for routine work will free up time for the judgment-heavy work that still requires a person. That is the actual opportunity on the table.