OpenAI reports a widening gap in enterprise agentic AI use
OpenAI says enterprises are moving from using AI for assistance to delegating work to agents, but adoption is becoming increasingly uneven. In June, firms in the top 10% of monthly AI use generated 8.3 times as many output tokens per active user as typical firms, compared with a 2.6-times gap in January.
The findings appear in two OpenAI publications: Enterprise Signals, which examines agentic AI across its enterprise customers, and the working paper How Organizations Use AI: Evidence from ChatGPT. OpenAI defines typical firms as those between the 45th and 55th percentiles of output tokens per active user.
Agentic work takes a larger share of enterprise usage
As of June, Codex generated 64% of combined Codex and ChatGPT output tokens among enterprise customers. OpenAI attributes the figure to both the frequency of Codex use and the longer, multi-step work associated with agentic workflows. Products such as ChatGPT Work and Codex can use tools, create files and produce material for human review.
This model differs from asking an assistant for advice. An employee may instead ask an agent to collect information from multiple sources and draft a presentation. The approach builds on the move toward a ChatGPT Work agent that can take tasks through to a result, described in ChatGPT Work agent for enterprise execution as an enterprise workflow designed for execution rather than answers alone.
Plugins and skills separate frontier firms
OpenAI reports that advanced capabilities are more common at firms with the deepest usage. Among weekly active users, 21% at frontier firms use Plugins and 19% use skills. The corresponding figures at typical firms are 9% and 3%.
Plugins can combine reusable instructions, called skills, with applications that connect an agent to company data, tools and actions. OpenAI gives the example of a sales Plugin combining a team playbook with CRM access so an agent can prepare a tailored response using current customer information and previous proposals. OpenAI says 95% of its own active users use Plugins weekly.
Growth extends beyond engineering
Codex adoption is also spreading across knowledge-work functions. Since February, weekly active enterprise Codex users increased 108 times in legal, 41 times in sales, 41 times in recruiting and 26 times in marketing. Engineering grew fivefold over the same period.
Virgin Atlantic is cited as an example. Its engineering teams use Codex to refactor legacy code in 30 minutes rather than two weeks, while product teams use ChatGPT Work to conduct weeks of competitive research in hours for its five-year digital strategy. Enterprise Signals draws on a sample of more than 10 million messages.
OpenAI also found that early-career employees use AI more intensively than senior staff. Six months after adoption, early-career employees sent 13 more messages per week than executives. For businesses, the practical implication is to identify effective employee workflows and standardise them with appropriate access controls, governance and human review.