TL;DR
OpenAI has published two studies of how its corporate customers use its products. The headline claim is a widening gap: the top tenth of firms by usage now consume 8.3 times as much model output per active user as median firms, against 2.6 times in January. Agent products account for most of that consumption, and the fastest growth is in functions with no engineering staff at all.
What the numbers say
By June, Codex was producing 64% of the combined output OpenAI’s enterprise accounts generated across Codex and ChatGPT — a shift from asking models things to handing them tasks. Growth by function since February looks dramatic: 108 times in legal, 41 times each in sales and recruiting, 26 times in marketing, against just 5 times in engineering, which had the head start.
The capability gap is more interesting than the volume gap. At the heaviest-using firms, 21% of weekly users touch Plugins and 19% use skills; at typical firms those figures are 9% and 3%. OpenAI notes its own staff sit at 95%. There is also a finding that cuts against most survey data: junior employees use the tools more than executives do, sending roughly 13 more messages a week six months in.
How much of this to believe
The measure carrying the headline is tokens per active user, which is a reasonable proxy for depth of use and an exact proxy for spend. A vendor reporting that its best customers buy more, and that this gap is widening, is describing its own revenue concentration as an adoption insight. The function-level multiples come off a February base close to zero. And the comparison group — a firm at the median of OpenAI’s own customer list — has already bought in, so none of this speaks to organisations that have not.
Looking forward
The UK example given is Virgin Atlantic, where engineering teams reportedly refactor legacy code in half an hour rather than a fortnight, and product teams compress weeks of competitor research into hours feeding a five-year digital plan. Treat the specific numbers as a customer’s own account. The transferable observation is where the growth sits: legal, sales, recruiting and marketing are precisely the functions where UK firms have the least mature governance, the fewest technical reviewers, and the most confidential material passing through.