OpenAI’s two agentic products have grown from roughly 2 million to 10 million combined weekly active users in about four months — and the curve is steepening, not flattening.
GROWTH TRAJECTORY — CODEX + CHATGPT WORK COMBINED WAU
Mid-March 2026
~2 million combined weekly active users — baseline at product launch phase
July 12, 2026
~6 million combined weekly active users — 3× growth over four months
July 21, 2026
~10 million combined weekly active users — OpenAI characterizes recent stretch as ~2× in a single week
According to Bloomberg, OpenAI disclosed that Codex — its coding agent — and ChatGPT Work have reached roughly 10 million combined weekly active users as of July 21, 2026. The trajectory is the story: approximately 2 million in mid-March, 6 million by July 12, and 10 million by July 21, with OpenAI characterizing the most recent stretch as roughly 2× growth in a single week. The hedges belong in the same sentence: this is an OpenAI-disclosed figure, the company has not clarified whether “active” means daily, weekly, or a point-in-time snapshot, and it discloses little about how deeply these products are actually being used.
Codex and ChatGPT Work sit in a distinct product category from the base ChatGPT chat interface. Both are agentic: they accept a task, take multi-step actions autonomously, and return a completed output rather than a single-turn answer. That distinction matters for the adoption read — users are not just querying, they are delegating. The curve describes a behavioral shift, not merely a feature adoption.
The key insight: A curve that steepens — rather than bends — after crossing 6 million is unusual. Most consumer software adoption decelerates as it exhausts early adopters. If the 10M figure holds and the definition of “active” is consistent, OpenAI’s agentic products are still in the accelerating phase of an S-curve, not the plateau.
The Structural Read
The Agentic AI Stack — a framework we mapped on Business Engineer — describes a fundamental reorientation of where AI value accretes: away from the model-as-interface layer (chatbots, search augmentation) and toward the model-as-executor layer (agents that own tasks end-to-end). What OpenAI’s numbers illustrate, with all their definitional caveats, is that this reorientation is now visible in user behavior at scale.
The same signal is arriving from Anthropic. Claude Code scaled from roughly $1 billion to $8 billion in annualized run-rate in six months — covered in detail here. Two frontier labs, independently, are watching their agentic products outrun everything else in the portfolio. That convergence is a more durable signal than either data point alone.
Agentic AI Stack — Structural Pattern
The chatbot is the discovery layer. The agent is the retention layer.
Chatbots acquire users; agents embed into workflows. Once a user delegates a repeating task to an agent, switching cost compounds with every completed job. The growth engine and the moat are the same product.
The honest bracket, stated plainly: self-reported weekly-active figures with an undefined “active” threshold are not the same as audited engagement metrics. A curve this steep will bend — the question is when and at what level. And adoption figures do not translate directly to durable revenue without knowing task depth, enterprise contract penetration, and churn. What they do signal is where the marginal user is choosing to go.
Three Implications
COMPETITIVE POSITIONING
OpenAI is signaling to enterprise buyers that agentic adoption is not experimental — it is happening at consumer and prosumer scale right now. That changes the procurement conversation: pilots become harder to justify when a vendor can cite 10 million weekly actives across two products.
INFRASTRUCTURE DEMAND
Agentic workloads are compute-heavier per session than single-turn queries — multi-step reasoning, tool calls, and longer context windows stack up. A move from 6M to 10M WAU in nine days implies a non-trivial inference demand spike that flows directly to the GPU and data-center stack beneath it.
MARKET STRUCTURE
When both frontier labs see agents outrun their chatbot products as growth drivers, the center of gravity in AI productstrategy shifts. Teams building on top of model APIs need to recalibrate: the interface layer that wins user time going forward is the one that takes work off the user’s plate, not the one that answers questions.
Business Engineer Framework
The Agentic AI Stack
The Agentic AI Stack maps the nine layers where AI value is forming — from infrastructure and model providers down to the agent-execution and workflow-embedding layers where Codex and ChatGPT Work now compete. Understanding which layer you occupy (or build on) determines your margin profile, switching-cost dynamics, and exposure to commoditization. Read the full framework on Business Engineer.
The 10 million figure is OpenAI-disclosed, “active” is undefined, and the curve will bend — but when two leading frontier labs independently report that their agentic products are outpacing everything else in the portfolio, the direction of value migration is no longer a thesis. It is a measurement.
Gennaro is the creator of FourWeekMBA, which reached about four million business people, comprising C-level executives, investors, analysts, product managers, and aspiring digital entrepreneurs in 2022 alone | He is also Director of Sales for a high-tech scaleup in the AI Industry | In 2012, Gennaro earned an International MBA with emphasis on Corporate Finance and Business Strategy.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
