How To Put OpenAI’s Agentic AI Numbers For EU Business In Perspective

🔍 Read the full analysis: How To Put OpenAI’s Agentic AI Numbers For EU Business In Perspective on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on monitors, keyboards and dev gear

• Business-only prices and quantity discounts
• Tax-exempt purchasing
• Multiple users, one account, clear invoices

Create a free Business account →Free to register

As an affiliate, we earn on qualifying purchases.Hide

OpenAI says its Codex coding agent and ChatGPT Work produced 55% of output tokens in its EU ChatGPT Business and Enterprise accounts by September 30, a figure that indicates growing use of agent products but does not show that agents handle most business tasks. The October 14 report also records an 8.1-fold rise in weekly active users of those products at businesses with 200 or fewer users since early May; it does not give absolute user counts.

OpenAI Economic Research’s report, Europe’s Agentic Opportunity: Moving from breadth to depth with AI agents, covers accounts in the 27 EU member states. OpenAI defines agentic output as tokens generated by Codex and ChatGPT Work. By September 30, Codex accounted for 34% of output tokens and ChatGPT Work for 21%; standard ChatGPT generated the remaining 45%. CFO Sarah Friar described the result as more than half of AI use among the company’s EU enterprise and business customers involving these tools.

The report says 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 October 5. Every member state recorded at least fivefold growth, with Belgium at 12.1 times, Italy at 10.8 and Luxembourg at 10.0. OpenAI does not provide the underlying country-level account or user counts, so these figures do not show how many businesses or workers the growth represents.

Agentic output shares also varied by country, from 85% in Lithuania and 81% in Cyprus to 22% in Slovenia and 26% in Austria. Germany’s share was 51%, ranking 17th among the 27 states. These are measures of output within OpenAI’s customer accounts, not national estimates of AI adoption.

What Token Growth Can Show

The figures offer a view into how OpenAI’s business customers are using its products, and they suggest that agent tools are generating a substantial share of model output in those accounts. That matters to companies tracking the move from conversational AI toward software that can work through files, browsers and company tools. But output volume is not the same as business impact: the data do not establish whether agents save time, improve work quality, reduce costs or produce revenue.

OpenAI itself cautions that tokens are an imperfect measure of business value and says agentic tasks can require substantially more computation. An agent may read documents, plan a response, generate code and check its work, producing much more text than a short chat exchange. The source article offers an illustrative calculation: if an agent run generated 10 to 30 times the output of a chat reply, a 55% token share could correspond to roughly 4% to 11% of interactions. That is an assumption, not a figure measured by OpenAI, but it shows why token share cannot be read as task share.

The adoption figures also carry a product-timing caveat. ChatGPT Work launched in July 2026, after the early-May starting point used for the small-business comparison. The report says a major part of the subsequent growth followed that launch. The result is evidence of uptake for OpenAI’s new offering among its customers; on its own, it does not establish a broad change across Europe’s small-business economy.

How OpenAI Defines Agent Use

OpenAI published the report through its Economic Research team alongside a Global Affairs post titled The EU’s next chapter in AI is agentic. Its global Enterprise Signals page provides related worldwide data. The figures in the EU report are drawn from ChatGPT Business and Enterprise accounts, not a representative survey of all European companies or a count of every AI system used in the region.

For the report, “agentic” refers specifically to Codex and ChatGPT Work. Codex is OpenAI’s coding agent; ChatGPT Work is an agent mode launched in July that can work with files, browsers and company tools. The report also compares so-called frontier firms: the top 10% by output tokens per active user. In September, EU frontier firms generated about four times as many tokens per user as typical EU firms. OpenAI says frontier firms outside the EU generated more than twice as many as the EU frontier, and that EU companies represented about 10% of frontier firms worldwide, compared with about 65% in North America.

Those comparisons describe OpenAI’s own account activity and its chosen definition of frontier use. They do not independently establish which companies are most advanced across the global economy, or why differences in output occur.

What the EU Figures Leave Out

The report does not disclose absolute account or user counts for the small-business growth measure or the country charts. Without those counts, readers cannot tell whether an eightfold increase began with a small number of users or a substantial base, or how much a few large customers influence a country’s share. OpenAI’s customer data also cannot show how common agent use is among businesses that do not use its products.

It remains unclear how much of the higher token volume reflects more agentic tasks, longer or more compute-intensive tasks, or changes in the mix of customers and products. The source article notes that ChatGPT Work contributed 21 percentage points of the September output share and that Codex rose from roughly one-fifth of output tokens in May to 34% by September. Those figures make product launches relevant to the trend, but they do not isolate the effects of each product or show whether usage will persist.

Nor do the country rankings measure national readiness or productivity. OpenAI reports no country-level sample sizes, and its figures may reflect the particular businesses in its customer base. The report does not provide independent evidence on realized savings, work quality or economic returns.

Evidence Needed Beyond Product Use

The next useful evidence would include absolute user and account totals, consistent measurements over longer periods, and clearer breakdowns by company size and industry. Those details could help distinguish sustained adoption from growth off a small base or a short-term response to a product launch. OpenAI has not, in the material summarized here, announced a timetable for publishing those figures.

To assess economic effects, businesses and researchers would need measures beyond tokens and active users: for example, how agent-assisted work compares with other workflows on time, cost, accuracy and quality. Until such evidence is available, the report is best read as a vendor-specific snapshot of product activity, not proof that agents now dominate work across European business.

Key Questions

What does OpenAI’s 55% figure measure?

It measures the share of output tokens generated by Codex and ChatGPT Work in OpenAI’s EU ChatGPT Business and Enterprise accounts by September 30, 2026. It is not a share of users, tasks or business value.

Did agent use grow 8.1 times at all EU small businesses?

No. OpenAI reports an 8.1-fold rise in weekly active users of Codex or ChatGPT Work among businesses with 200 or fewer users in its accounts, comparing the first week of May with the week ending October 5. It does not give absolute user totals, and the figures do not cover all EU small businesses.

Why can token share overstate how common agent use is?

Agent runs can generate much more text than ordinary chat exchanges because they may read files, plan, produce work and check results. A large share of tokens can therefore come from a smaller share of interactions. The report does not state the share of interactions that were agentic.

Do the country rankings show which EU nations are most AI-ready?

No. They show agentic output in OpenAI’s customer accounts by country. OpenAI does not provide country-level sample sizes, and the results may reflect differences in which customers use its services rather than national adoption overall.

Source: ThorstenMeyerAI.com