OpenAI reports that 98% of its staff are currently utilizing Codex agents, although all the information is based on self-reporting.

      TL;DR: OpenAI claims that 98% of its employees now use Codex agents, with non-developer usage increasing 137 times, though all metrics are self-reported by the company selling the product.

      According to a paper released on Wednesday titled “The Shift to Agentic AI: Evidence from Codex,” nearly 98% of OpenAI’s workforce utilizes Codex, the company’s AI coding agent, rising from about 40% in August 2025. This paper outlines a fundamental shift in how the company’s employees engage with AI, evolving from using conversational chatbots to autonomous agents performing multi-step tasks. However, all statistics in the document originate from OpenAI itself, which has a financial motive to promote the product being measured.

      The reported figures are noteworthy. Active users of Codex increased fivefold during the first half of 2026, while task requests expected to take eight hours or longer rose almost tenfold. The legal team at OpenAI generated 13 times more tokens in June compared to November 2025, a statistic that the company cites as proof that agents are spreading to various departments beyond engineering.

      The surge in non-developer use is central to OpenAI's story. The usage of Codex by individual non-developers has increased 137 times since August 2025, while organizational non-developer use has risen 189 times, and internal non-developer adoption has grown twelvefold. Earlier this month, the company expanded Codex with enterprise plugins linking 62 business applications, and non-developers now account for about 20% of the platform's five million weekly users, adopting it three times quicker than engineers.

      The paper presents this as evidence of a market transition from chatbots to agents. OpenAI contends that the patterns observed within its own operations, where departments such as legal and recruiting consider Codex an essential tool, foreshadow a wider enterprise AI adoption. The company cites external data indicating Codex usage among organizations stands at around 17% and among individuals at less than 1%, implying significant growth potential.

      However, the disparity between internal and external usage raises concerns about the representativeness of OpenAI’s workforce. The paper does not specify whether the company incentivizes or encourages employee use of Codex, an important point given that nearly universal adoption within a company selling the product does not equate to organic demand. No independent third party has verified the reported usage statistics.

      The issue of self-reporting also affects productivity claims, as OpenAI asserts that the increase in longer task requests and higher token generation demonstrates agents managing more complex tasks. As noted by The Register, faster code generation does not necessarily lead to proportional productivity improvements, as verification, testing, and deployment times may extend to accommodate faster output. The paper lacks data on whether the transition to agents has measurably enhanced output quality or decreased overall completion time.

      The wider context involves a competitive race among AI companies to demonstrate that agents, rather than chatbots, signify the next market phase. In May, OpenAI combined ChatGPT and Codex under Greg Brockman, streamlining its product strategy around a single agentic platform ahead of a potential Q4 IPO. Competitors like Anthropic’s Claude Code and Google’s Gemini are pursuing similar strategies, intensifying the pressure to showcase adoption growth.

      Meta’s internal experience serves as a similar reference point. The company launched a “Claudeonomics” leaderboard in April to track token consumption by teams, making AI usage a visible performance measure. This method, like OpenAI’s paper, focuses on input volume rather than output value, which is significant when the companies reporting the figures are also those selling the tools.

      The paper may be most informative as an indication of OpenAI's vision for the market and what it wants investors to perceive before the IPO. The shift from chatbot queries to autonomous agent tasks is indeed occurring and is evident across the industry. Whether it is progressing at the pace and scale that OpenAI's self-reported data suggests is a question that will only be answered through independent measurement.