Enterprise AI may be entering a new phase of inequality
The first wave of business AI adoption was defined by access. Which companies had deployed the tools? How many employees had seats? Were workers experimenting with chat interfaces at all? OpenAI’s new B2B Signals research suggests those questions no longer capture the frontier. The emerging divide, according to the report, is not merely whether firms use AI, but how deeply they use it inside day-to-day work.
The headline figure is striking. Frontier firms, defined as those at the 95th percentile of usage, now use 3.5 times as much intelligence per worker as typical firms, up from 2 times a year earlier in April 2025. OpenAI frames that measure using tokens generated as a proxy for the amount of work employees are asking AI to do. Tokens are not presented as a direct measure of value, but as a way to estimate the depth of AI use.
The report’s key argument is that the advantage is compounding. Once companies move beyond broad access and into more complex, production-oriented use, they appear to widen the distance from peers who are still treating AI mainly as a lightweight assistant.
Why message volume is not the whole story
One of the more consequential claims in the report is that message volume explains only 36% of the frontier advantage. In other words, the gap is not just that leading firms are asking AI more questions. It is that they are asking for richer, more complex work, providing more context, and generating more substantive outputs.
That distinction matters because it changes how enterprise adoption should be evaluated. A company can report growing activity and still remain relatively shallow in its usage. If employees are relying on AI only for simple prompts or occasional drafting help, the organization may not be capturing the kinds of workflow transformation that drive a stronger competitive edge.
OpenAI’s framing suggests that depth is becoming the more relevant metric. Firms at the frontier appear to be integrating AI into actual processes rather than treating it as an auxiliary convenience. That is a harder transition because it requires governance, enablement, and workflow design, not just software access.
Delegated work is becoming a marker of the frontier
The report singles out advanced tools and agentic workflows as the area where the strongest differences show up. Frontier firms, OpenAI says, send 16 times as many Codex messages per worker as typical firms. That is one of the clearest signals in the dataset that the next stage of enterprise AI may center on delegated work rather than simple chat assistance.
This is an important shift in how business AI maturity is described. If early enterprise adoption was about employees asking for help, frontier adoption increasingly looks like employees assigning work. That moves AI from the category of productivity layer into something closer to operational infrastructure. It also raises the stakes for implementation, because delegated workflows typically require tighter controls, clearer success criteria, and stronger trust in outputs.
OpenAI says leading firms are building governance for production use, investing in enablement, measuring depth, scaling what works, and moving from chat-based assistance to delegated workflows with agents. The pattern it describes is organizational, not just technical. The frontier is not defined only by tool choice, but by whether a company can turn AI use into a managed system rather than a scattered habit.
How the data was presented
B2B Signals is described as a recurring measure of how AI is diffusing across businesses, based on privacy-preserving, aggregated signals from enterprise use of OpenAI products. The company says the analyses rely on de-identified, aggregated enterprise usage data. It also says message content was classified using automated systems and that no OpenAI employee reviewed individual enterprise, business, or API customer data as part of the analysis.
Those methodological notes matter because enterprise AI reporting often runs into a trust problem. Companies want benchmarking and adoption signals, but they also want confidence that sensitive business use is not being manually inspected. OpenAI’s emphasis on aggregation and de-identification is meant to address that tension while still supporting a broad view of usage patterns.
At the same time, readers should note what the report is and is not claiming. It presents tokens as a proxy for demanded intelligence, not a direct measure of business outcomes. That means the findings are best understood as indicators of intensity and maturity of use, rather than proof that every extra token translates cleanly into productivity or profit.
What this means for enterprise strategy
If OpenAI’s findings hold more broadly, many companies may be benchmarking themselves against the wrong milestone. The differentiator is no longer basic deployment. It is whether AI is embedded deeply enough to handle complex, context-rich, delegated tasks in live workflows.
That has practical implications. Firms that want to move closer to the frontier may need to invest less in expanding access for its own sake and more in training, governance, process redesign, and selection of tasks that can sustain deeper use. The report suggests that advanced adoption does not happen automatically after rollout. It has to be cultivated.
It also hints at a feedback loop. Companies that learn to use AI deeply may discover new tasks worth delegating, which in turn encourages more infrastructure, more governance, and more internal expertise. That is the compounding advantage the report points to. Once a firm begins to operationalize AI at depth, its lead may become harder for slower adopters to close.
The emerging competitive gap
The broadest takeaway is that enterprise AI competition may now be entering a more structural phase. The difference between leaders and typical firms is no longer simply experimentation versus non-experimentation. It is the difference between shallow usage and integrated, delegated, production-grade use.
That is a more serious divide because it touches capability, not just adoption optics. A firm with deep AI use may not merely work faster on familiar tasks. It may reorganize how work is assigned and executed. If that pattern continues, the frontier advantage OpenAI describes could become less like a temporary head start and more like a durable operating model.
For enterprises still treating AI as a broad but light productivity tool, the message is blunt. Access was the first milestone. Depth may be the one that determines who actually pulls ahead.
This article is based on reporting by OpenAI. Read the original article.
Originally published on openai.com
