On August 12, 2026, OpenAI published two complementary studies: Enterprise Signals and How Organizations Use AI: Evidence from ChatGPT. Their central argument is that enterprise AI is moving from helping people think toward executing parts of the work, but organizations are making that transition at very different speeds.
OpenAI calls the top 10% of firms by monthly AI-use intensity frontier firms and uses output tokens per active user as a proxy for depth of use. As of June, those firms generated 8.3 times as many output tokens per active user as typical firms, up from a 2.6 times gap in January. This is not a direct measure of productivity or revenue. It is a signal that some companies are using longer, more contextual, and more multi-step agentic workflows.
Codex is another part of the shift. OpenAI says that, as of June, Codex generated 64% of combined Codex and ChatGPT output tokens among enterprise customers. The share reflects both how often Codex is used and how much output its tasks produce, so it should not be read as 64% of enterprise work being fully automated. It is better understood as evidence that agent-style work is taking on longer assignments.
Frontier firms also use plugins and skills more often. Among weekly active users, 21% at frontier firms use plugins compared with 9% at typical firms; skills use is 19% versus 3%. OpenAI also reports that 95% of its own employees use plugins weekly. These are OpenAI customer and internal usage analyses, not a guarantee for other companies, but the direction is clear: connecting AI to company context, tools, and reusable instructions is closer to deep adoption than simply opening a chat window.
Agentic use is also spreading beyond engineering. OpenAI reports that weekly active enterprise Codex users grew 108 times in legal, 41 times in sales, 41 times in recruiting, and 26 times in marketing since February, compared with 5 times in engineering. Growth multiples depend heavily on the starting base and do not describe the total market size of each function. They do show that agent workflows are no longer only a coding-team story.
The research also reports a result that differs from common survey intuition: six months after adoption, early-career employees sent 13 more messages per week than executives. OpenAI's practical implication is to identify employees with strong AI habits and turn effective individual workflows into shared methods, rather than measuring adoption only by whether senior leaders say they use AI.
The most useful operating framework in the article has three parts: give agents the context and tools required to complete valuable work; establish clear permissions, human review, and governance; and turn successful individual experiments into shared workflows. That is a synthesis of the research rather than a separate OpenAI metric. Teams moving from assistance to execution should measure success rate, human-edit rate, exceptions, and cost before deciding which steps an agent may handle and which still require approval.
