OpenAI Enterprise Signals: Frontier Firms Pull 8.3x Ahead in AI Adoption

On August 12, 2026, OpenAI published its most detailed enterprise adoption study to date, and the headline number is stark: frontier firms — the top 10% of AI users by monthly volume — now generate 8.3 times as many output tokens per active user as typical enterprises. That figure stood at 2.6x in January. In six months, the gap has more than tripled.

The report, titled “From Assistance to Execution: How Enterprises Put AI to Work,” draws on aggregated usage data from OpenAI’s entire enterprise customer base, paired with a companion academic working paper published on arXiv (“The Shift to Agentic AI: Evidence from Codex”). Together, they paint the clearest picture yet of how AI is actually being used inside companies — not how executives say it’s being used, but what the token logs show.

The 8.3x Gap: What’s Driving It

The core finding is that AI adoption is bifurcating. A small group of frontier firms is pulling dramatically ahead while the majority remain in early-stage experimentation. As of June 2026, Codex — OpenAI’s agentic coding and workflow tool — generated 64% of combined Codex and ChatGPT output tokens among enterprise customers. Because agentic workflows execute longer, multi-step tasks, they inherently produce far more output per session than a typical chatbot conversation.

This represents what OpenAI calls the shift “from assistance to execution.” Users who once asked AI how to prepare a presentation are now asking agents to gather information from multiple sources and draft the presentation itself. The model isn’t just answering questions; it’s doing work.

Message volume alone, however, doesn’t explain the gap. OpenAI’s earlier B2B Signals report found that message volume accounts for only 36% of the 3.5x difference observed in May. The remaining 64% comes from depth — how deeply AI is embedded into actual workflows, how many tools it connects to, and how many repeatable processes it automates.

Departments Beyond Engineering Are Catching Up Fast

Perhaps the most surprising data in the report concerns which departments are growing their AI usage fastest. Since February 2026, weekly active enterprise Codex users grew at dramatically different rates depending on the function:

• Legal: 108x growth
• Sales / Account Management: 41x
• Recruiting: 41x
• Marketing / Communications: 26x
• Healthcare / Clinical: 24x
• Finance / Accounting: 20x
• Engineering: 5x

Engineering’s 5x growth is the slowest of any tracked function — but only because engineers were already the heaviest Codex users six months ago. The explosive growth is happening in departments that had near-zero agentic AI usage in February. Legal teams, in particular, went from barely using Codex to becoming one of the fastest-growing segments, suggesting applications in contract review, compliance analysis, and legal research are gaining real traction.

Virgin Atlantic serves as a case study in the report. Its engineering teams use Codex to refactor legacy code in 30 minutes instead of two weeks, while product teams compress weeks of competitive research into hours using ChatGPT Work. Richard Masters, VP of Data and AI at Virgin Atlantic, noted: “The trajectory of Codex is thinking beyond pure engineers. It’s moving into a real tool for everyone.”

Integration Is the Real Differentiator

What separates frontier firms from typical firms isn’t just how much they use AI — it’s how they connect it to their existing systems. Frontier firms are significantly more likely to use Plugins and Skills, both of which link AI agents to company-specific data sources, tools, and repeatable processes:

• Plugins: 21% of weekly active users at frontier firms vs. 9% at typical firms
• Skills: 19% vs. 3%

For context, 95% of OpenAI’s own employees use Plugins weekly. The gap in Skills adoption — 19% versus 3% — is particularly striking. Skills allow organizations to define custom, repeatable AI workflows that encode institutional knowledge. Frontier firms aren’t just chatting with AI more; they’ve systematically built AI into the fabric of how work gets done.

Critically, the frontier advantage holds across industries and company sizes. Intensive AI use isn’t limited to tech companies. Healthcare, finance, and traditional enterprise organizations are represented among the frontier cohort, suggesting that organizational strategy — not sector — determines who pulls ahead.

Junior Employees Out-AI Their Bosses

One finding contradicts nearly every industry survey on AI adoption: six months after deployment, early-career employees use AI more per week than executives. Most prior research, based on self-reported survey data, has reported higher usage among senior leaders and managers. OpenAI’s usage logs tell a different story.

The implication is significant. Survey data may systematically overstate executive AI usage — executives may feel pressured to report adoption they haven’t fully internalized — while understating the grassroots adoption happening among junior staff who quietly integrate AI into daily tasks. For leaders, OpenAI’s recommendation is to identify employees with the strongest AI habits, who may be the most junior members of the organization, and make their workflows visible so effective practices spread across all levels.

The Cost Question OpenAI Doesn’t Answer

What the reports deliberately don’t address is cost. The data shows a correlation between deep AI adoption and stronger financial measures — enterprise adopters tend to hold more assets, employ more workers, and invest more in R&D. But correlation is not causation. The reports arrive at a moment when enterprise AI spending faces intense scrutiny.

The cautionary tales are real. Uber reportedly burned through its entire 2026 AI budget in four months after rolling out coding tools to roughly 5,000 engineers, with its COO telling Fortune it was difficult to connect rising AI costs to measurable product improvements. Gartner forecasts AI agent software spending will reach nearly $207 billion in 2026, up 139% from $86.4 billion in 2025 — yet few AI pilots have made it to production.

OpenAI’s data makes a compelling case that frontier firms are pulling ahead. It doesn’t yet prove that the spending required to get there pays for itself.

What This Means for Enterprises

The signal from OpenAI’s report is unambiguous: the gap between AI leaders and laggards is widening, and it’s widening fast. Organizations that have already built AI deeply into their workflows are compounding their advantage — more integrations, more Skills, more agentic workflows that generate exponentially more output.

For companies still treating AI as an experimental side project, the window to catch up is narrowing. The data suggests that the path to the frontier isn’t about buying more AI — it’s about integrating AI more deeply into fewer, higher-impact workflows, connecting agents to real systems and data, and institutionalizing the practices that already work, regardless of where in the org chart they originate.

The next six months will determine whether the 8.3x gap continues to widen — or whether the cost reality finally catches up with the ambition.