The AI Usage Gap Between Top-Tier and Typical Companies Tripled in the First Half of the Year

Three reports, the same day, the same signal

The coincidence is real, not manufactured. On August 12, OpenAI published the current edition of Enterprise Signals, based on its own telemetry from its enterprise customer base. The same day, Deloitte published a survey of 501 US leaders — from senior manager to C-suite, all at least piloting agentic AI — and Liferay published a separate survey of 500 US professionals involved in AI decisions, run through the third-party platform Pollfish.

Three organizations, three different methodologies — proprietary usage telemetry, an executive-perception survey, a market survey — three different populations, published independently, without citing each other. But all three describe the same rift: adoption is deepening fast at a core of companies, while organizational readiness — process, governance, measurement — lags well behind for the rest.

All four sources in this article (the fourth, PwC, is from April) also have an adjacent commercial interest — OpenAI sells exactly what it measures, Liferay sells AI governance tools, Deloitte and PwC sell AI transformation consulting. The methodologies are public and verifiable, but it's worth saying outright, not glossing over.

The gap that tripled: 2.6× → 8.3×

OpenAI splits its enterprise customers into "frontier" firms — the top 10% each month by output tokens per active user — and "typical" firms — the 45th-55th percentile. It isn't comparing the top to the whole market, but to the middle of the distribution.

In January 2026, frontier firms generated 2.6 times as many tokens per active user as typical firms. By June, that ratio had climbed to 8.3×. OpenAI itself calls this "a threefold increase" in the gap — an acceleration, not a linear continuation of a trend.

The gap varies by industry: 11.7× in Information/Technology, the highest, versus 5.3× in Manufacturing, the lowest. Typical firms grow modestly everywhere — between 1.9× and 2.8× over the past year. The difference doesn't come from typical firms falling behind. It comes from frontier firms pulling further ahead.

OpenAI itself puts a limit on the number: "tokens are an imperfect measure of business value — a short response can be highly valuable, while a long one may add little. But token volume offers a useful proxy for the depth of AI use." In other words: the gap shows how much work a company is delegating to an agent, not how much value it's extracting from it. Related, but not identical.

What's clear in the data: agentic use (Codex) generated 64% of combined Codex+ChatGPT output tokens at enterprise customers in June — a sign that "doing" has overtaken "asking." And the growth is no longer coming from engineering alone: since February, weekly active enterprise Codex users have grown 108× in legal, 41× in sales, 41× in recruiting, 26× in marketing — versus just 5× in engineering.

Adoption is outrunning readiness

If the OpenAI gap shows WHAT is happening, Deloitte's survey suggests why typical firms are falling behind — without directly explaining the OpenAI gap; these are different samples.

42% of the leaders Deloitte surveyed say their business has tested or deployed AI agents, and 43% have deployed them across more than one function. But only 15% have reached orchestrated, multi-agent adoption at scale. Even among that 15%, only 46% believe their processes are actually ready for it. Only 5% call themselves "highly prepared."

Across the seven readiness areas Deloitte measured, the weakest is exactly the one that matters most for scaling: business processes, at 21%. Governance, risk, and security sit at 39%. The barriers cited are concrete: 72% say they lack unified, accessible data, 70% say they can't yet trust or govern agents enough, 67% say integration is too costly or complex.

Liferay finds a similar pattern in a different sample: 54% of companies are already running AI agents in production or actively piloting them — not a population stuck in testing; most have something functional. But only 25% measure impact with clear KPIs, and only 24% have an AI usage policy that applies company-wide.

The two surveys don't prove causation — they don't show that a lack of governance "produces" the usage gap OpenAI measured. They show something just as useful: across different samples, companies are simultaneously reporting fast adoption and organizational readiness that hasn't kept up. It's a convergent symptom, not a proven mechanism.

The signal isn't new. The difference can now be tracked over time

This theme isn't new to massai.ro readers. In July we wrote about the McKinsey report showing that fewer than 10% of companies scale AI agents to tangible value — a figure that has, in fact, circulated in McKinsey's research for over a year, since June 2025.

What's new in August isn't the finding itself, but the fact that it can now be tracked from several angles at once. An April precedent reinforces this: PwC's study of 1,217 executives across 25 sectors, published on April 13, already showed that 20% of companies capture nearly 74% of AI's economic value — leaders 2.6× more likely to say AI improves their ability to reinvent their business model, 2.8× more likely to increase the number of decisions made without human intervention, 1.5-1.7× more likely to have a formal AI governance framework. PwC's conclusion back in April: "without a shift in approach, the performance gap between AI leaders and laggards is likely to widen further."

Four months later, that's exactly what the August 12 cluster shows — this time with fresh numbers, from three simultaneous sources.

What this means for your company

None of the four reports proves definitively why the gap exists. But all four, independently, point to where the weak spots are — exactly the spots a company controls, regardless of which AI model it uses.

The useful question is no longer "do we have AI in the company" — across surveys of 500 to 1,200 respondents, most companies already have something, even if informal. The questions that matter are more concrete: which workflows have actually been redesigned around an agent, not just had AI "layered on" top of an old process? (Deloitte finds that most companies are still just layering AI onto existing processes rather than redesigning them — only 1 in 5 say they're ready for full redesign.) What KPI are you tracking for each active agent, and who owns the outcome? Is there a mechanism for a successful use in one team to become standard practice across the rest of the company, or does it stay isolated?

The gap OpenAI measured isn't a prophecy. It's a snapshot of what happens when some companies are already answering these questions, and the rest aren't yet.

Sources: ↗ OpenAI — Enterprise Signals · ↗ OpenAI — From assistance to execution · ↗ Deloitte — AI Agents are Only the Beginning · ↗ Liferay — 2026 Agentic AI Adoption and Governance Report · ↗ PwC — AI Performance Study · ↗ MassAI — McKinsey: AI agents don't scale past 10%