OpenAI Reports Most EU Business AI Is Agentic—Read The Fine Print

🔍 Read the full analysis: OpenAI Reports Most EU Business AI Is Agentic—Read The Fine Print on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on office and shipping supplies

• 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 reported that its agent products generated 55% of output tokens in its ChatGPT Enterprise and Business accounts across the European Union by September 30, a measure of how much text its systems produced, not how many tasks or users relied on agents. The finding, published October 14, offers a snapshot of adoption among OpenAI’s EU business customers, while leaving open how representative that activity is of the wider European economy.

OpenAI Economic Research’s report, Europe’s Agentic Opportunity: Moving from breadth to depth with AI agents, divides agentic output between Codex, at 34%, and ChatGPT Work, at 21%. The remaining 45% came from plain ChatGPT, according to the report. OpenAI defines agentic use here as work with Codex, its coding agent, or ChatGPT Work, an agent mode introduced in July 2026 that can use files, browsers and company tools.

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 said every EU member state recorded at least fivefold growth over that period. Belgium led at 12.1 times, followed by Italy at 10.8 and Luxembourg at 10.0. The report does not provide the underlying user totals.

Reported agentic output shares varied by country, from 85% in Lithuania and 81% in Cyprus to 22% in Slovenia and 26% in Austria. Germany’s share was 51%, placing it 17th among the 27 member states. These are figures for OpenAI’s customer accounts, not national statistics covering all businesses or AI systems.

OpenAI also compared its highest-use businesses, defined as the top 10% by output tokens per active user. In September, EU frontier firms produced about four times as many tokens per user as typical EU firms. Frontier firms outside the EU produced more than twice the output per user of the EU frontier group. OpenAI estimates EU companies account for about 10% of frontier firms worldwide, compared with about 65% in North America.

What the Token Share Measures

The headline number indicates that agent products account for a large share of generated text within the OpenAI business accounts examined. That matters to companies evaluating how quickly agent features are entering workplace workflows, and to policymakers weighing whether adoption is translating into more capable business use.

But output tokens are not a count of completed tasks, active workers, or business gains. OpenAI’s Enterprise Signals page describes tokens as an imperfect measure of business value and says agentic tasks can require substantially more computation. An agent may read files, plan, generate code or documents, check its work and try again, producing far more text than a short chat response. The 55% figure therefore cannot establish that most AI interactions are agentic or that agents deliver most of the value.

The distinction also matters when interpreting growth. ChatGPT Work launched in July, while the report compares May with early October. OpenAI’s chart shows the product contributed 21 percentage points of output by September 30. Its introduction accounts for more than half of the roughly 35-point rise in the agentic share since May, based on the chart figures. That is evidence of rapid uptake of a new product among OpenAI customers, not by itself proof of a lasting shift across European businesses.

How OpenAI Defines Agent Use

The report covers ChatGPT Enterprise and ChatGPT Business accounts in the EU’s 27 member states. Its definition of agentic use is limited to two OpenAI products: Codex and ChatGPT Work. The reported output shares are measured through September 30, while the small-business weekly-active-user comparison ends October 5.

OpenAI’s policy message is that Europe should move from broad adoption to deeper use of AI agents, through worker skills and confidence and wider access to advanced capabilities. The report’s comparisons suggest that, within OpenAI’s own customer data, the most intensive EU users still trail similar firms outside the bloc. They do not establish the performance or adoption levels of all European companies.

OpenAI CFO Sarah Friar said more than half of AI use among the company’s EU enterprise and business customers “now involves these tools.” The report’s more specific 55% statistic is a share of output tokens. That wording should not be read as saying more than half of all customers, employees or work tasks use agents.

Limits of the EU Customer Data

OpenAI does not publish country-level account or user counts in the report. That makes it difficult to judge how much a small number of large customers may affect country shares, particularly in smaller markets. The company itself notes that smaller countries such as Cyprus, Estonia and Lithuania have relatively high agentic use within their larger firms.

The report also gives no absolute user totals for the 8.1-times small-business growth figure, so the starting base and the number of businesses represented cannot be assessed from the published multiplier. Because the comparison begins before ChatGPT Work launched, product availability is part of the period being measured. OpenAI says a major part of the growth followed that launch.

It remains unclear how these account-level measures translate into the number of agent-assisted tasks, time saved, productivity or revenue. The supplied report material also does not establish whether OpenAI’s customer sample represents the wider EU business population or how usage compares with other AI providers. The country rankings should be read as patterns in OpenAI’s accounts, not as a ranking of national AI readiness.

What Further Evidence Would Show

OpenAI has not announced a follow-up date or a planned release of underlying account counts in the material available. The immediate next step for readers assessing the claims is to distinguish the report’s measured activity—output tokens and weekly active users—from outcomes it does not report, such as business productivity or task completion.

More detail on sample sizes, absolute user numbers and the mix of company sizes would help assess the country and small-business comparisons. Independent studies tracking tasks and business outcomes over time would also show whether the increase reflects sustained changes in work or mainly use of newly launched OpenAI products. Until such evidence is available, the report is best treated as a vendor-reported snapshot of activity among its EU business customers.

Key Questions

What does OpenAI mean by “55% of output tokens”?

It means Codex and ChatGPT Work generated 55% of the text tokens produced in the covered EU ChatGPT Enterprise and Business accounts by September 30. It does not mean 55% of users or tasks used agents.

Which products did OpenAI count as agentic?

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

Did small-business agent use rise 8.1 times across all EU businesses?

No. OpenAI reports an 8.1-times increase in weekly active users of Codex or ChatGPT Work among businesses with 200 or fewer users in its accounts. The comparison runs from the first week of May to the week ending October 5, and the report does not give absolute user totals.

Does the report show that agents improved productivity?

No. It reports product usage and output tokens, not measured productivity, revenue or time savings. OpenAI says tokens are an imperfect measure of business value.

Do the country figures represent all businesses in each EU country?

No. They describe OpenAI’s covered customer accounts. The report does not provide country-level sample sizes, so the figures should not be treated as national adoption rates or a ranking of each country’s overall AI readiness.

Source: ThorstenMeyerAI.com