Why OpenAI’s EU Agentic AI Stat Needs A Denominator Check

🔍 Read the full analysis: Why OpenAI’s EU Agentic AI Stat Needs A Denominator Check on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get smart everyday buys delivered free with Prime

• Fast, free delivery on millions of items
• Prime Video, Amazon Music and more included
• Member-only deals all year

Start your free Prime trial →Free trial for eligible customers · Cancel anytime

As an affiliate, we earn on qualifying purchases.Hide

OpenAI said AI agents generated 55% of output tokens in its ChatGPT Enterprise and Business accounts across the 27 EU member states by September 30, according to a report published October 14. The figure marks growing use of the company’s agent products, but it is a share of generated text—not a measure of how many users or tasks involve agents, or how much business value they create.

OpenAI Economic Research’s report, Europe’s Agentic Opportunity: Moving from Breadth to Depth with AI Agents, counts output from Codex, OpenAI’s coding agent, and ChatGPT Work, its agent mode for working with files, browsers and company tools. OpenAI says ChatGPT Work launched in July 2026. By September 30, Codex accounted for 34% of output tokens in the covered accounts and ChatGPT Work for 21%; standard ChatGPT accounted for the remaining 45%.

The report also says weekly active users of Codex or ChatGPT Work at businesses with 200 or fewer users rose 8.1 times from the first week of May to the week ending October 5. OpenAI reports at least fivefold growth in every EU member state, with Belgium at 12.1 times, Italy at 10.8 times and Luxembourg at 10 times. It does not provide the underlying absolute user counts.

OpenAI’s country figures vary considerably: agentic output ranges from 85% in Lithuania to 22% in Slovenia. Germany is at 51%, which the report ranks 17th among the 27 countries. The report also defines “frontier firms” as the top 10% of firms by output tokens per active user. It says EU frontier firms generated about four times the output per user of typical EU firms, while frontier firms outside the EU generated more than twice as much as the EU frontier.

What the Token Share Measures

The 55% figure offers a view of where computation is going within OpenAI’s business accounts. It does not mean that most people, work assignments or company AI use in Europe have shifted to agents. An agent can read files, plan, generate a response and revise it, producing far more tokens than a short chat exchange. A token share can therefore rise faster than the share of interactions involving agents.

OpenAI’s own Enterprise Signals material calls tokens an imperfect measure of business value and says agentic tasks can require substantially more computation. Those qualifications matter for policymakers and businesses weighing whether European firms are adopting AI effectively. Token volume alone cannot establish productivity gains, cost savings or improved work quality.

The distinction also matters when comparing adoption across countries. Without account and user counts for each country, the reported percentages cannot show whether a national result reflects broad uptake or a small number of especially active customers. The figures describe OpenAI’s observed customer base, not necessarily the whole European economy.

Product Launch Shapes the Trend

The report’s adoption window begins in early May, before ChatGPT Work launched in July. OpenAI’s chart shows Codex rising from roughly one-fifth of EU output tokens in May to 34% by September 30; ChatGPT Work then contributed 21 percentage points. That means a substantial part of the agentic share reflects a newly introduced product, rather than a long-running measure of changing work practices.

OpenAI says a major part of the small-business user growth followed ChatGPT Work’s introduction. The 8.1-times figure compares weekly active users at two points in time, but the report gives no starting or ending user totals. Growth from a small base can produce a large multiplier. The report’s policy argument is that Europe should turn broad AI adoption into deeper use through worker skills, confidence and more accessible advanced capabilities; the adoption statistics alone do not establish that those outcomes have occurred.

Missing Counts Limit Comparisons

The report does not disclose absolute user totals, account counts by country or the number of businesses behind each national percentage. That makes it difficult to judge how representative country comparisons are, particularly for smaller markets. OpenAI notes that smaller countries including Cyprus, Estonia and Lithuania have relatively high agentic use within their larger firms.

The data also leave open how much of the increase reflects durable changes in workplace behavior, as opposed to the launch and growing availability of ChatGPT Work. The report does not show whether agent use produced measurable gains in productivity, revenue or work quality. Nor does the 55% token share provide an estimate of the share of interactions that were agentic. Any conversion from tokens to tasks would require assumptions not supplied by OpenAI.

Evidence Needed Beyond Token Growth

OpenAI’s report points to deeper business use, worker skills and access to advanced tools as areas for further progress. To evaluate whether adoption is broadening, future data would need to provide absolute user and account counts, clear definitions for active use, and repeated measurements over time that separate product-launch effects from continued uptake.

For claims about economic impact, token data would need to be paired with evidence about task completion, quality, time saved and costs. Until those measures are available, the report is best read as a snapshot of agent activity in OpenAI’s own EU business accounts—not a complete measure of agent adoption or business value across Europe.

Key Questions

What does OpenAI’s 55% figure measure?

It measures the share of output tokens generated by Codex and ChatGPT Work in the covered EU ChatGPT Enterprise and Business accounts by September 30, 2026. It is not the share of users, tasks or business value attributable to agents.

How did OpenAI define agentic tools?

The report counts Codex, OpenAI’s coding agent, and ChatGPT Work, an agent mode that can work with files, browsers and company tools.

What does the 8.1-times small-business increase mean?

OpenAI says weekly active users of Codex or ChatGPT Work at businesses with 200 or fewer users increased 8.1 times between the first week of May and the week ending October 5. The report gives no absolute user totals, and says a major part of the growth followed ChatGPT Work’s launch.

Does the report show that agents improved company productivity?

No. It reports usage measures, including output tokens and weekly active users, but does not establish productivity gains, cost savings or improved work quality.

Source: ThorstenMeyerAI.com