OpenAI's agent push has crossed 10 million users, but the number only tells you one thing for certain: developer tools are turning into workplace platforms.
Ten million is a number worth sitting with. Bloomberg reported on July 21 that OpenAI now has 10 million people using Codex and ChatGPT Work, nearly twice the level from earlier in July, when ChatGPT Work was announced. That is the story. But don't read it as a clean weekly-active Codex number, because OpenAI has not given that split or that active-user window for the combined total.
The cleaner benchmark came from OpenAI itself on June 2, when the company said Codex had more than 5 million weekly active users, up more than six-fold since the Codex desktop app launched in February. OpenAI also said about 20% of Codex users were knowledge workers rather than developers. Then ChatGPT Work arrived on July 9 with Codex built in, plus tools for creating finished documents, spreadsheets, presentations, reports and Sites. The count moved fast after that. Posts from OpenAI's Codex lead Tibo Sottiaux pointed to 6 million, 7 million and 8 million combined active users across Codex and ChatGPT Work between July 12 and July 14, according to Codex Insider's review of those milestones.
That caveat matters. A founder, CIO or engineering lead can't treat 10 million combined users as the same thing as 10 million paid enterprise seats. OpenAI hasn't disclosed how many users belong to Codex versus Work, how many are on paid plans, or how often they run agents. The figure is still useful. It tells you demand is no longer theoretical. It does not tell you revenue quality.
The coding market changed underneath OpenAI
The market Codex is chasing looked very different eighteen months ago. GitHub Copilot had the enterprise relationships and Microsoft distribution. Cursor, built by Anysphere, had developer affection and extraordinary revenue growth. Forbes reported that Cursor had topped $4 billion in annualized revenue in early June, up from $3 billion in late April and $2 billion in February. That is not normal SaaS growth. It's a product people were already paying for before the market finished naming the category.
Then SpaceX agreed to buy Anysphere for $60 billion in an all-stock deal announced June 16. Reuters reported the deal, and SpaceX's own SEC filing said the merger was expected to close in the third quarter of 2026, subject to approvals. Cursor is still a real product with real revenue. But being folded into Elon Musk's SpaceX changes the question for buyers. If you're making a three-year tooling decision, you now have to ask what Cursor looks like inside a rocket, satellite and AI infrastructure company.
GitHub is moving too. In April, GitHub said all Copilot plans would transition to usage-based billing on June 1, with usage measured through GitHub AI Credits and calculated from token consumption. That is a blunt signal. Microsoft wants Copilot priced more like infrastructure than a flat software seat. Anthropic's Claude Code and Claude Cowork add another pressure point, especially for teams that already trust Claude for long-context technical work.
The market is four-way now. Not one leader with stragglers.
What founders should actually take from the number
Here's the thing about user milestones in enterprise software: they're a leading indicator, not a business model. OpenAI's edge is not only that Codex has scale. It's that ChatGPT Work pulls Codex into a broader workflow, where a user can ask for research, a spreadsheet, a report or a web app and keep the agent running across a longer task. OpenAI's July 9 release described Work as an agent that can gather information across apps and workflows and turn goals into finished materials. That is a bigger ambition than code completion.
Costs will show it. If those 10 million users mostly run light tasks inside existing subscriptions, the milestone is impressive but financially thinner than it looks. If a meaningful share starts burning paid usage on long agent runs, OpenAI has something closer to a new enterprise consumption business. You don't need to guess which version matters more.
For operators, the practical question is simple. Are your engineers already using Codex, Cursor, Copilot or Claude Code without a company standard? If they are, pretending this is still an experimental category is a waste of time. JetBrains said in April that 90% of professional developers in its January 2026 AI Pulse survey regularly used at least one AI tool at work for coding and development tasks. Stack Overflow's 2025 survey was more cautious, finding 84% using or planning to use AI tools, but the direction is the same.
Don't skip the boring step. Pick the workflow before you pick the logo. Codex and ChatGPT Work are built around async, sandboxed agents that can work across files and produce finished artifacts. Cursor is still strongest when the editor is the center of gravity. Copilot is tied tightly to GitHub and Microsoft's enterprise stack. Claude Code has become a serious default for engineers who want a terminal-first agent and are willing to steer it closely.
OpenAI's 10 million number does not settle that choice. It does make one thing plain: the coding wars are no longer only about who writes the next function fastest. They are about who owns the place where work gets broken into tasks, delegated to agents, reviewed by humans and billed by usage. That is where the budget is moving.
Also read: Google released three Gemini models in one day while its flagship is still stuck in testing • Intel cuts jobs in its fastest-growing division two days before Q2 earnings • Supermicro books $60 billion in orders in a single quarter as AI server backlog hits record high
