OpenAI says Codex and ChatGPT Work now have 10 million users combined, a sharp jump that shows how quickly the company is trying to turn ChatGPT from a chatbot into a workplace agent platform.
According to Bloomberg reporting summarised, OpenAI says usage of Codex and ChatGPT Work has nearly doubled from earlier this month, when ChatGPT Work was announced. Unite.AI also reported that OpenAI shared the 10 million figure ahead of a planned July 21 announcement, but did not break out how many users belong to Codex versus ChatGPT Work.
That caveat matters. A combined number is still impressive, but it is not the same as saying 10 million people are using a single product in the same way. Codex is a coding agent. ChatGPT Work is a broader task agent for office workflows. Together, they show OpenAI’s direction: agents that do work, not only answer questions.
The number says OpenAI has momentum in agentic software. It also says the company understands how important enterprise adoption has become. Consumer ChatGPT made OpenAI famous, but workplace agents may be where the larger recurring revenue sits.
Codex already had traction with developers, and OpenAI previously said it had passed millions of weekly active users. ChatGPT Work gives the same agent idea a wider audience: people who need documents, spreadsheets, research, slide drafts, app workflows and business tasks handled across tools.
The big question is how sticky the usage will be once launch promotions, usage-limit resets and curiosity fade. A user opening ChatGPT Work during a launch week is not the same as a company standardising on it for daily operations.
OpenAI is no longer only competing in chat. It is competing for the software layer of work. Microsoft has Copilot, Anthropic has Claude Code and Claude-style workplace tools, Google has Gemini inside Workspace, and startups are building vertical agents for law, finance, sales, support and engineering.
That makes ChatGPT Work strategically important. If OpenAI can become the agent layer that sits across apps, files and workflows, it can create a business that is harder to replace than a chatbot tab. Users do not only ask questions; they delegate tasks.
But that also raises trust questions. Agents that run for long periods, touch files and act across tools need guardrails, audit logs and approval steps. Recent AI security stories, including the Hugging Face agentic AI breach, show why companies will be cautious about handing too much autonomy to software.
The most interesting part of the Codex story is that OpenAI wants it to matter beyond traditional software engineering. If a non-developer can ask an agent to build a dashboard, clean a spreadsheet, write a script or connect business tools, the addressable market becomes much larger.
That is also why the line between Codex and ChatGPT Work is blurring. Coding agents can become workflow agents because many business processes are really small software problems hiding behind forms, documents and spreadsheets.
This is where the competition with Anthropic becomes sharp. Claude Code has become popular with developers, while Claude’s long-context strengths appeal to knowledge workers. OpenAI is trying to answer with one unified agent surface inside ChatGPT.
The 10 million figure is useful, but it should be read carefully. OpenAI has not clearly defined the activity window, the split between products or the number of paying business seats. Those details matter if we want to understand whether this is durable enterprise adoption or a surge around a launch.
Still, the direction is clear. The AI market is moving from chat interfaces to agents that act. OpenAI wants Codex and ChatGPT Work to become the tools people keep open while real work is happening.
If the company can prove reliability, security and usefulness at scale, the workplace may become the next major front in the AI race. If it cannot, the 10 million number will be remembered as a strong launch signal rather than a long-term platform shift.