The Denominator Changes How To Read OpenAI’s EU Agentic AI Claim

🔍 Read the full analysis: The Denominator Changes How To Read OpenAI’s EU Agentic AI Claim on ThorstenMeyerAI.com

Age 18–24?Offer from Amazon

Prime made for students and young adults

• Fast, free delivery for dorm and study essentials
• Prime Video and Amazon Music included
• Member-only deals

Try Prime for Young Adults →Free trial for eligible 18–24 year olds

As an affiliate, we earn on qualifying purchases.Hide

OpenAI said Codex and ChatGPT Work generated 55% of output tokens in its ChatGPT Enterprise and Business accounts across the European Union by Sept. 30, 2026, a sign that agent products now account for more of the model output in those accounts than standard ChatGPT. The figure measures tokens, not tasks, users or business value, so it does not establish that most AI use in European businesses is agentic.

The figures appeared in OpenAI’s October 14 report, Europe’s Agentic Opportunity: Moving from breadth to depth with AI agents, published by OpenAI Economic Research alongside a Global Affairs post. The report covers ChatGPT Enterprise and Business accounts in the 27 EU member states. It defines agentic use as activity by Codex, OpenAI’s coding agent, and ChatGPT Work, an agent mode launched in July that can work with files, browsers and company tools.

OpenAI’s reported 55% share comprises 34% from Codex and 21% from ChatGPT Work as of Sept. 30. Standard ChatGPT accounted for the other 45%. The company also said weekly active users of the two agent products at businesses with 200 or fewer users rose 8.1 times between the first week of May and the week ending Oct. 5. Every member state recorded at least fivefold growth over that period, according to the report. OpenAI did not provide absolute user counts.

The report also shows wide variation across countries and company sizes. Agentic output’s share ranged from 85% in Lithuania and 81% in Cyprus to 22% in Slovenia and 26% in Austria; Germany was at 51%. OpenAI defines “frontier firms” as the top 10% of companies by output tokens per active user. In September, EU frontier firms generated about four times the tokens per user of typical EU firms, while frontier firms outside the EU generated more than twice as many as the EU frontier. OpenAI said EU companies represented about 10% of frontier firms worldwide, compared with about 65% in North America.

What the Token Share Measures

The 55% figure is a measure of where generated tokens went inside OpenAI’s business accounts. It is not a count of how many workers use agents, how many workplace tasks agents perform, or the value those tasks produce. OpenAI’s own Enterprise Signals page describes tokens as an imperfect measure of business value and says agentic tasks can require substantially more computation.

That distinction matters because an agent may read files, plan, generate material and check or revise its work, producing far more tokens than a short chat response. The report does not give the average output per interaction for either category, so the token split cannot be translated into a share of interactions. The 55% headline supports the narrower conclusion that agents account for most of the output tokens in these particular accounts—not that they account for most AI use across European business.

The data also point to a gap OpenAI wants policymakers and businesses to address: while some European firms use its products heavily, the company’s comparison places EU frontier customers behind their counterparts elsewhere. OpenAI frames the policy challenge as turning broader adoption into deeper use, through worker skills, confidence and access to advanced tools. That is the company’s interpretation of its customer data, not an independent measure of Europe’s overall AI capacity.

A New Product Shapes the Trend

The reported growth period starts in early May, before ChatGPT Work launched in July 2026. OpenAI’s chart shows Codex at roughly a fifth of EU output tokens in May and 34% by Sept. 30; ChatGPT Work then contributed 21 percentage points. The article accompanying the report says the new product accounts for more than half of the roughly 35-point increase in agentic token share since May. That makes the result partly a measure of adoption of a newly available product, rather than a clean view of a longer-term change in workplace habits.

The same timing affects the 8.1-times small-business growth figure: it compares weekly active users against a May baseline and includes the period after ChatGPT Work’s introduction. OpenAI says a major part of the increase followed that launch. Without absolute user numbers or a longer history, the multiplier shows sharp growth from the stated baseline but does not reveal how many businesses or workers use the products, or whether the pattern will persist.

Country rankings also need caution. The report gives no account or user counts by member state and notes that smaller countries can have high agentic use within their larger firms. The figures are therefore best read as a snapshot of OpenAI’s customer base, not as a ranking of national readiness. The report’s two country measures do not move together neatly: the source analysis says Slovenia had the lowest agentic share but above-median small-business growth, while Lithuania had the highest share but below-median growth.

What the EU Figures Cannot Show

OpenAI has not published country-level account counts, user counts or absolute small-business adoption figures in the material described here. That limits how confidently readers can interpret national comparisons, especially for smaller markets, and makes it impossible to judge the scale behind the 8.1-times growth rate. The report also does not supply a breakdown of agentic tasks by industry or explain how many tokens a typical agent run generates relative to a chat interaction.

The data cover customers of OpenAI’s paid business products, not all EU companies or workers. They do not independently establish how widespread agents are across the European economy, whether agent use improves productivity, or whether the gap between EU and non-EU frontier firms reflects AI capability, customer mix, spending, task choice or other factors. The company’s own analysis is informative about its accounts, but it is not a representative survey of European businesses.

It is also unclear whether the reported growth will continue after the initial period following ChatGPT Work’s launch. OpenAI attributes a major part of small-business growth to the product’s introduction, but the report does not isolate the effects of that launch from other causes. Further data over a longer period would be needed to assess whether the rise reflects sustained adoption or a launch-related surge.

More Data Needed on Adoption

The next useful measure would be a follow-up that reports absolute active-user and account totals, alongside token volumes, across a longer period. That would help readers distinguish rapid growth from a small baseline from widespread adoption and show whether ChatGPT Work’s initial contribution persists. Country comparisons would be easier to assess with sample sizes and information about customer composition.

For now, OpenAI’s report provides a dated snapshot through Sept. 30 for output shares and through the week ending Oct. 5 for small-business user growth. The company’s stated next challenge is moving from broad adoption toward deeper use. Whether that means more routine agent use, measurable productivity gains or a larger share of business tasks is not answered by the published figures.

Key Questions

What does OpenAI mean by agents generating 55% of output?

It means Codex and ChatGPT Work produced 55% of output tokens in OpenAI’s covered EU Enterprise and Business accounts by Sept. 30, 2026. It does not mean agents handled 55% of tasks or generated 55% of business value.

Which products count as agentic in the report?

OpenAI counts Codex, its coding agent, and ChatGPT Work, an agent mode launched in July 2026 that can work with files, browsers and company tools.

How fast did agent use grow at small businesses?

OpenAI reported that 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 Oct. 5, 2026. The report gives no absolute user total for that comparison.

Does the report show that European businesses are behind?

OpenAI’s comparison says EU frontier firms produced fewer tokens per user than frontier firms outside the EU in September 2026. That is a comparison within OpenAI’s customer data, not a comprehensive measure of all European companies or their productivity.

What remains unknown about the country figures?

The published material does not provide account or user sample sizes by country. Without those counts, the rankings may reflect differences in OpenAI’s customer mix as well as differences in agent use across markets.

Source: ThorstenMeyerAI.com