There is an analysis that enterprise artificial intelligence (AI) is evolving beyond the support stage of simply answering questions to the execution and delegation stages in which it directly performs complex tasks.
OpenAI on Aug. 13 released a report, "Enterprise Signals," analyzing how global corporations are using AI.
According to the report, while conventional AI assistants were limited to assisting with individual tasks such as information search or document drafting, Agentic AI connects directly with internal tools and computer systems at corporations to autonomously, or under supervision, execute complex assignments such as file edits and multi-step processes.
OpenAI's survey found that as of June, Codex, an Agentic AI product, accounted for 64% of total output tokens from corporate customers. Even factoring in the heavy computation typical of complex, time-consuming work, OpenAI interpreted this as an indicator that the purpose of AI adoption by corporations is shifting decisively toward real task delegation.
A "frontier gap" between corporations is also widening rapidly depending on their ability to use AI, it was found. According to the report, "frontier corporations," the top 10% by monthly usage, generated 8.3 times more output tokens per active user than mid-tier general corporations. The gap, which was 2.6 times on Jan. 1, expanded more than threefold in just half a year. By industry, OpenAI said the usage gap between frontier corporations and general corporations was largest in information technology (IT) at 11.7 times.
OpenAI analyzed that this gap stems not from simple model access, but from the use of advanced features such as "skills" and "plugins" that enable consolidation of organizational context, data and internal tools. Among weekly active users at frontier corporations, plugin usage was 21%, more than double that of general corporations (9%), and skill usage at frontier corporations (19%) likewise far outpaced general corporations (3%). Citing that 95% of its internal staff use plugins weekly, OpenAI said deep integration with organizational systems is the key factor that determines AI's real value.
The report also tallied that Agentic AI is rapidly spreading beyond traditional development teams to knowledge work overall. Since Feb., weekly active Codex users have surged 108-fold in legal, 41-fold each in sales and recruiting, and 26-fold in marketing, far outpacing growth in engineering (fivefold).
An analysis of sample messages exceeding 10 million found that while ChatGPT is used mainly for writing and communication, Agentic AI messages were 75% focused on coding and system/agent operations. OpenAI assessed this to mean Agentic AI is taking hold as a practical team-level tool for executing real work.
By rank, lower-tenure employees showed the highest AI usage, with usage declining as rank rose, indicating that productivity advantages from AI are pronounced among lower-tenure workers, the report found.