Why The Denominator Matters In OpenAI’s EU Agentic AI Claim

🔍 Read the full analysis: Why The Denominator Matters In OpenAI’s EU Agentic AI Claim on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get hardware and tech essentials 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 Codex and ChatGPT Work generated 55% of output tokens in its ChatGPT Enterprise and Business accounts in the 27 EU member states by Sept. 30, 2026. The figure, published in a report on Oct. 14, describes the volume of text generated by OpenAI’s agents—not the proportion of users, business tasks or economic value attributable to them.

OpenAI’s Economic Research report, Europe’s Agentic Opportunity: Moving from Breadth to Depth with AI Agents, defines agentic use as activity from Codex, its coding agent, and ChatGPT Work, an agent mode launched in July that can work with files, browsers and company tools. In the reported EU accounts, Codex produced 34% of output tokens and ChatGPT Work 21%; ordinary ChatGPT produced the remaining 45%.

The company also reported that weekly active users of Codex or ChatGPT Work at businesses with 200 or fewer users rose 8.1 times between the first week of May and the week ending Oct. 5. OpenAI said every EU member state recorded at least fivefold growth. It did not provide absolute user counts in the material described in the report.

OpenAI’s country figures show agentic output shares ranging from 85% in Lithuania and 81% in Cyprus to 22% in Slovenia and 26% in Austria. Germany was at 51%, ranking 17th among the 27 member states. The report also says EU frontier firms—defined as the top 10% by output tokens per active user—generated about four times as many tokens per user as typical EU firms in September. Frontier firms outside the EU generated more than twice the output per user of the EU frontier.

What the 55% Figure Actually Measures

The denominator changes the meaning of the headline. Output-token share is not task share: an agent can read documents, plan, generate a response and revise it, using more tokens than a short chat exchange. OpenAI itself cautions that tokens are an imperfect measure of business value and says agentic tasks can require substantially more computation.

That means the 55% figure supports a limited but useful conclusion: in the covered OpenAI business accounts, agents generated more of the model’s output than standard ChatGPT did. It does not establish that agents handled most work, served most users or created most value. The report does not provide the task counts or economic outcomes needed to answer those questions.

The growth figures also need their time window. ChatGPT Work was introduced in July, after the May starting point for the small-business comparison. OpenAI says a major part of the increase followed that launch. Rapid adoption may indicate that customers are trying the new feature, but the figures alone do not show whether use will persist or produce measurable gains in productivity.

A New Agent Mode Shapes the Trend

OpenAI published the report alongside its Global Affairs post, The EU’s Next Chapter in AI Is Agentic, on Oct. 14. The figures cover OpenAI’s own paid business products, not all European companies or the wider AI market. The global Enterprise Signals page offers a broader comparison, but the EU report’s account-level observations remain specific to the company’s customer base.

The product timeline matters when reading the reported rise. OpenAI’s chart, as described in the source material, shows Codex at roughly one-fifth of EU output tokens in May and 34% by Sept. 30. ChatGPT Work was not available until July, then accounted for 21% by the end of September. The report’s policy argument is that Europe should convert broad AI adoption into deeper use by building skills and confidence and making advanced capabilities more accessible and affordable.

OpenAI also reports a gap between its EU frontier firms and leading firms outside the bloc. It estimates that EU companies represent about 10% of frontier firms worldwide, compared with about 65% in North America. These comparisons reflect OpenAI’s definition of frontier firms and its own customer data; they are not a comprehensive census of companies in either region.

Missing Counts Limit Country Comparisons

OpenAI’s published material, as summarized in the source, does not give account or user sample sizes for each country. That makes it difficult to know how representative national percentages are, especially in smaller markets where a small number of large customers could affect the result. The country figures should not be treated as rankings of national AI readiness.

The report also does not provide absolute weekly active-user counts for the small-business growth measure, a count of agentic tasks, or independent measures of business outcomes. Without those denominators and outcome data, readers cannot tell how much of the increase reflects a larger active customer base, more intensive use by existing customers, or the launch of ChatGPT Work. The analysis in the source material notes that country rankings for agentic share and small-business growth do not closely align; that comparison is not itself an OpenAI measure.

It remains unclear whether the reported usage levels will continue after the initial launch period, how much value customers obtain from the tools, and whether the gap between EU and non-EU frontier firms reflects adoption, customer mix, industry differences or other factors. OpenAI’s data describes its own accounts, so it cannot establish the pattern across all European businesses.

Evidence Needed Beyond Token Volume

The next useful evidence would include absolute user and account counts, country-level sample sizes, and consistent time-series data that separates product-launch effects from sustained adoption. Reporting task counts alongside tokens would help show whether agents are handling more work or simply producing longer outputs.

To assess the report’s productivity argument, companies and researchers would also need outcome measures—such as time saved, quality, costs and changes in completed work—rather than usage volume alone. OpenAI has not supplied those measures in the material covered here. Its report sets out a policy case for training and wider access, while the durability and economic impact of agent use remain open questions.

Key Questions

What does OpenAI’s 55% EU agent figure measure?

It is the share of output tokens generated by Codex and ChatGPT Work in OpenAI’s ChatGPT Enterprise and Business accounts across the 27 EU member states by Sept. 30, 2026.

Does 55% mean agents handled most business tasks?

No. The report’s figure measures generated text, not the number of users, interactions or tasks. Agent runs can use more tokens than ordinary chat, and the report does not provide task counts to establish their share of work.

Why did small-business agent use rise 8.1 times?

OpenAI measured weekly active users of Codex or ChatGPT Work at businesses with 200 or fewer users from the first week of May to the week ending Oct. 5. The company says a major part of the growth followed ChatGPT Work’s July launch. It did not provide absolute user counts in the source material.

Do the country percentages rank national AI adoption?

Not reliably. They describe agentic output in OpenAI’s accounts, and the report does not provide country-level sample sizes. The figures are not a census of businesses or a direct measure of national AI readiness.

Does the report show that agents improve productivity?

It documents usage patterns, but the figures cited do not establish productivity gains or business value. The report does not provide independent outcome measures such as time saved, work quality or cost changes.

Source: ThorstenMeyerAI.com