A company can learn a lot about the market by looking at its own employees. OpenAI says that its team members are switching from chatbots to agents as their primary form of AI interaction, a trend also detected (though less pronounced) among external organizations and users. Instead of one-off ChatGPT prompts, workers are asking Codex agents to tackle multi-step tasks that take long periods of time. And those doing so are increasingly non-developers. OpenAI insists that its findings have implications for other companies, labor researchers, and policymakers, not the least of which would be a brighter revenue picture for OpenAI. Longer running tasks consume more tokens, and to the extent those can be billed, that should help diminish hundreds of billions in debt obligations. "We find that agentic AI usage is growing rapidly: the number of active users has grown more than fivefold in the first half of 2026, with the most rapid increase occurring outside the initial audience of software developers," said company researchers and academics in a paper [PDF] titled, "The Shift to Agentic AI: Evidence from Codex." OpenAI did not immediately respond to a request to clarify whether it incentivizes or encourages employees to use its AI tools – through internal communiques, token allocations, token use leaderboards, or tying tool usage to performance metrics. But we'll take it on faith that when there's enough Kool-Aid on-premises, employees may just develop a taste for it regardless of whether their jobs depend on Kool-Aid consumption. "Through August 2025, the average OpenAI worker spent less than 10 percent of their tokens on Codex," the biz explained in a blog post accompanying its paper (that suggests employee token allocations). "Now, every department, including non-technical departments such as Legal and Recruiting, uses Codex as their primary AI tool for work." Within OpenAI, 97.9 percent of employees are now using Codex, up from around 40 percent in August 2025. External organizations have also seen a usage uptick, to 17.3 percent presently. With individuals, Codex isn't much to speak of – about 0.7 percent. The thing about Codex is that, as an agent, it can operate for long periods of time. "Since the start of the year, the share of individual Codex users who submit at least one request for a task estimated to require more than eight hours for an experienced human to complete has increased nearly tenfold," the paper says. We note that comparing the time a human might take for a task (as estimated by an LLM-as-judge) to the time an AI model takes is only part of the picture if the workflow isn't entirely automated. Generating code at, say, 10x the rate a person might manage may expand the time required for code verification and deployment. OpenAI also points out that, since August 2025, non-developer usage of Codex has risen 137x for individuals, 189x for organizational users, and 12x within OpenAI. The company concedes that technical usage remains the dominant mode, but insists that adoption by non-devs shows how a broader set of knowledge workers can take on coding or technical tasks, such as automation, data transformation, and data analysis. "In June 2026, the median OpenAI employee in a legal role generated 13 times more monthly output tokens across Codex and ChatGPT than they did in November 2025," the paper says. Given that the number of US federal lawsuits filed against OpenAI and associated entities only grew about 11 percent (35 to 39) between the last six months of 2025 and the first six months of 2026, it looks like OpenAI's legal team, with its 13x token surge, is making the company's case for the productivity benefits of AI tools. ®
