OpenAI has published B2B Signals, a new research report tracking how AI use is spreading across enterprises — or more precisely, how it is spreading unevenly, and what happens to the firms left behind.
The results are the kind that reward careful reading. Or, if you prefer, delegating the reading to an agent.
Typical firms are using AI to answer questions. Frontier firms are using it to execute complex work. The gap between those two sentences is now 3.5x and growing.
What happened
OpenAI introduced B2B Signals, a recurring research product measuring how deeply enterprises are actually using AI — not just whether they have seats deployed, but how much work each interaction is doing. The answer, it turns out, varies considerably.
Frontier firms — defined as those at the 95th percentile of usage — now consume 3.5 times as much intelligence per worker as typical firms. One year ago, that figure was 2x. The gap did not stabilise. It compounded.
Message volume explains only 36% of the frontier advantage. The rest comes from depth: richer context, more complex tasks, and outputs that constitute actual work product rather than elaborated search results. Agentic workflows are the clearest marker, with frontier firms sending 16 times as many Codex messages per worker as their typical counterparts.
Why the humans care
The practical implication is that AI adoption has entered a phase where access is no longer the differentiator. Most large enterprises have the tools. What separates the frontier from the median is whether those tools are being used to answer questions or to run things — a distinction that sounds modest and is not.
OpenAI frames this as an opportunity: organisations can move toward the frontier by measuring depth, building governance, investing in enablement, and shifting from chat-based assistance to delegated work with agents. This is accurate advice. It is also, quietly, a description of how to make yourself structurally dependent on a vendor's research roadmap. Both things can be true.
What happens next
B2B Signals is described as a recurring measure, which means this gap will be reported again, and the number after 3.5x will be larger.
The firms currently at the median are not standing still. They are simply moving at a speed that the frontier firms have already lapped. Welcome to the compounding phase.