OpenAI Data Shows AI Agents Drive More Than Half of EU Business Output Tokens

OpenAI says its agentic tools, Codex and ChatGPT Work, now generate more than half of output tokens among ChatGPT Enterprise and Business users in the European Union. The finding, published in OpenAI's Economic Research report on Europe's agentic AI adoption, points to a change in how organisations use generative AI: from primarily asking questions to assigning tools work that can move through planning and completion.

The strongest signal is not simply growing AI usage. It is the growing share of work associated with agentic products. In OpenAI's EU sample, weekly active users of Codex and ChatGPT Work at small and medium businesses grew more than eightfold from May 2026. OpenAI defines those businesses as accounts with 200 or fewer associated users, and says the growth occurred in every EU member state.

That makes the report relevant to teams assessing where AI can have a practical operational effect. But its figures should be read carefully. They measure output tokens within OpenAI's ChatGPT Enterprise and Business customer base, not all AI activity or every business in the EU. Token volume can indicate intensity of use, but it does not by itself prove a specific productivity gain, cost saving, or return on investment.

What OpenAI's EU data shows

OpenAI's analysis frames agentic AI as software that can work across a task rather than only provide a response to an individual prompt. Codex and ChatGPT Work account for more than half of the output tokens generated by the company's EU Enterprise and Business users, according to the report. This is a meaningful shift in the mix of activity inside OpenAI's business products.

Three findings stand out:

• Agentic usage is substantial: Codex and ChatGPT Work produce more than half of output tokens in the EU sample.
• Smaller-business adoption is accelerating: weekly active users of those tools among EU businesses with 200 or fewer associated users rose more than eightfold since May 2026.
• Adoption intensity differs sharply by country: the agentic share of output tokens ranges from roughly 22% in Slovenia to roughly 85% in Lithuania.

OpenAI defines EU frontier firms as the top 10% of firms by output tokens per active user. As of September, those firms generated more than four times as many output tokens per active user as typical EU firms. Yet OpenAI also found that EU frontier firms remained behind global frontier firms on this measure. The implication is that Europe has highly engaged adopters, but there is still a gap between the region's most intensive users and the leading firms globally.

The variation between countries deserves caution. A higher share of output tokens from Codex and ChatGPT Work does not automatically mean a country is more productive or has achieved better AI outcomes. Product mix, account composition, work patterns, and the timing of usage can all affect the metric. It does, however, show that the move toward agentic workflows is not occurring at an identical pace across the EU.

What businesses should take from the shift

For a business, the useful distinction is between experimentation and a workflow that produces a repeatable result. A conventional chatbot interaction might help someone draft an email or explain a spreadsheet formula. An agentic workflow is more relevant when a task has several connected stages, such as gathering information, preparing a first draft, checking it against defined inputs, and handing the result to a person for approval.

OpenAI's data does not identify which workflows deliver the best returns. It does support a more grounded conclusion: businesses are increasingly using its agentic products at a scale that goes beyond occasional prompt-based assistance. Leaders should therefore evaluate AI opportunities in terms of specific processes rather than broad claims about transformation.

A practical starting point is to identify work that has clear inputs, a repeatable sequence, and a measurable outcome. Suitable candidates may include internal research preparation, document processing, software development tasks, or recurring customer and operational work. Before expanding use, teams should establish:

• the task owner and the person responsible for reviewing results;
• the systems, documents, or data needed to complete the task;
• the quality threshold that makes the output useful; and
• a baseline for time, cost, errors, or throughput that can be compared after implementation.

This approach helps separate activity from value. High token use may reflect deep adoption, but ROI requires evidence that a workflow is faster, more reliable, less manual, or able to support more work without proportionate additional effort. It also makes workforce enablement important. OpenAI CFO Sarah Friar said Europe is moving toward AI-enabled productivity and highlighted the need to give workers the skills and confidence to use AI effectively.

For businesses considering adoption, the report is best treated as a directional benchmark rather than a market-wide scorecard. Its dataset is limited to OpenAI's Enterprise and Business accounts, and its definitions and product usage may not map neatly to another provider's customer base. Still, the trend is clear within that ecosystem: agentic tools are becoming a central part of business AI usage in the EU.

For companies that want to turn this interest into measurable operational gains, Scalevise can help design and implement AI workflow automation for recurring business processes. A focused automation project can connect the right inputs, define review steps, and measure whether a workflow reduces manual effort or improves throughput. Rather than deploying AI as a general-purpose experiment, build around a task with a clear owner and outcome. Request an AI automation consultation to identify the highest-value workflow to automate.

Frequently Asked Questions

What did OpenAI find about AI agents in EU businesses?

OpenAI found that Codex and ChatGPT Work generated more than half of output tokens among ChatGPT Enterprise and Business users in its EU sample. The company presents this as evidence of a shift toward agentic, task-oriented AI use.

How quickly is agentic AI use growing among EU small and medium businesses?

OpenAI says weekly active users of Codex and ChatGPT Work among EU businesses with 200 or fewer associated users grew more than eightfold since May 2026, with growth in every EU member state.

Does OpenAI's report prove that AI agents increase productivity?

No. The report measures output tokens within OpenAI's Enterprise and Business accounts. It signals usage intensity, but it does not directly measure productivity, cost savings, or ROI for individual businesses.

Why do AI agent adoption figures vary across EU countries?

OpenAI reports a range from about 22% of output tokens in Slovenia to about 85% in Lithuania for Codex and ChatGPT Work. The report's data do not establish a single cause, and product mix, account composition, and usage timing can affect comparisons.

Conclusion

OpenAI's EU data shows that agentic tools have moved into the mainstream of usage within its business customer base, while adoption among smaller organisations is rising quickly. The next question is not whether a company uses AI, but whether it can apply AI to defined workflows and measure the result. Businesses that pair practical implementation with worker skills and clear performance measures will be better placed to turn growing agent use into durable value.