OpenAI states that 98% of its workforce currently utilizes Codex agents, although the data is based on self-reporting.

      TL;DR: OpenAI reports that 98% of its employees are using Codex agents, with non-developer usage increasing by 137 times, but all metrics are based on the company's own self-reports.

      According to a paper released by OpenAI on Wednesday, titled “The Shift to Agentic AI: Evidence from Codex,” nearly 98% of OpenAI employees now utilize Codex, the company’s AI coding agent, a rise from about 40% in August 2025. The paper highlights a significant change in how the workforce engages with AI, shifting from using conversational chatbots to employing autonomous agents for multi-step tasks. However, all statistics presented in the paper come directly from OpenAI, a company that stands to benefit financially from promoting its own product.

      The highlighted figures are striking. There was a fivefold increase in active Codex users in the first half of 2026, and requests for tasks estimated to require eight hours or more surged almost tenfold. OpenAI's legal team generated tokens at a rate 13 times higher in June compared to November 2025, a statistic that the company claims reflects the widespread adoption of agents across various departments beyond just engineering.

      The growth in non-developer usage is central to OpenAI’s narrative. Individual non-developer engagement with Codex skyrocketed 137 times since August 2025, while organizational non-developer usage increased 189 times, and internal non-developer uptake grew twelvefold. Earlier this month, OpenAI expanded Codex with enterprise plugins that link to 62 business applications, and non-developers now account for approximately 20% of the platform's five million weekly users, adopting it three times quicker than engineers.

      The paper characterizes this as evidence of a broader market transition from chatbots to agents. OpenAI posits that the patterns observed within its own organization, where departments including legal and recruiting now consider Codex an essential tool, foreshadow a wider enterprise AI adoption on a similar trajectory. External data showing Codex usage among organizations at about 17% and among individuals at less than 1% suggests significant potential for growth.

      However, the disparity between internal and external adoption raises questions about the representativeness of OpenAI’s workforce. The paper does not clarify whether the company incentivizes or promotes the use of Codex among employees, a pertinent oversight since widespread adoption in a company selling the product does not equate to genuine market demand. No independent third party has verified any usage statistics.

      The problem of self-reporting extends to productivity claims as well, where OpenAI asserts that longer task requests and increased token generation indicate that agents are taking on more complex work. As noted by The Register, quicker code generation doesn't necessarily correlate with proportional productivity improvements, as the time required for verification, testing, and deployment may increase to utilize any speed gains. The paper does not provide data demonstrating whether the transition to agents has notably enhanced output quality or shortened overall completion times.

      The broader context involves a competitive race among AI companies to demonstrate that agents, rather than chatbots, represent the next evolution in the market. OpenAI unified ChatGPT and Codex under Greg Brockman in May, aligning its product strategy with a singular agentic platform ahead of a possible IPO in Q4. Similar agent-focused strategies are being pursued by Anthropic’s Claude Code and Google’s Gemini, heightening competitive pressure to showcase growth in adoption.

      Meta’s internal practices provide a comparable reference point. The company introduced a “Claudeonomics” leaderboard in April to track token usage by team, thus visualizing AI engagement as a performance indicator. Like OpenAI’s report, this method measures input rather than output, a crucial distinction when the reporting entities also sell these tools.

      Ultimately, the paper serves as an indicator of OpenAI’s perspective on market trends and aims to shape investor perceptions before its IPO. The genuine transition from chatbot interactions to autonomous agent tasks is evident across the industry. However, whether this is happening at the speed and scale suggested by OpenAI’s self-reported figures remains a question that only independent assessment can address.