OpenAI says EU small-business AI agent use grew more than eightfold
OpenAI said on 14 October 2026 that weekly active users of its Codex and ChatGPT Work tools among EU businesses with 200 or fewer users grew more than eightfold between early May and 5 October 2026. The company's economic research team also reported that these agent tools produce more than half of output tokens in its EU sample of ChatGPT Enterprise customers, with country shares ranging from 22% in Slovenia to 85% in Lithuania. The figures come from OpenAI's own customer accounts, not a survey of European firms, and they measure generated output rather than completed work. For businesses that automate routine tasks, the report signals where staff time is moving, not proof of productivity gains.
In short
- OpenAI reported that weekly use of its agent tools by EU businesses with 200 or fewer users grew more than eightfold from May to early October 2026.
- Agent tools make up more than half of output tokens in OpenAI's EU enterprise sample, from 22% in Slovenia to 85% in Lithuania.
- The figures come from OpenAI's own customer accounts and measure generated output, not completed work or productivity.
- Businesses handing tasks to agents need clear rules on access, review and who signs off the result.
What OpenAI reported
OpenAI published a report from its Economic Research team on 14 October 2026, titled "The EU's next chapter in AI is agentic". The company says agent tools, which carry a task from planning through to completion rather than simply answering a question, now make up a large share of the work its EU business customers do with its software. According to OpenAI, weekly active users of Codex and ChatGPT Work among EU businesses with 200 or fewer users rose more than eightfold between early May and the week ending 5 October 2026. The company says this growth appeared in every member state.
OpenAI defines a small or medium business in this report as one with 200 or fewer associated users. Codex is the company's coding agent. ChatGPT Work is an agent mode that the company launched in July 2026 and that can work with files, browsers and company tools. The report also covers larger accounts, where OpenAI says Codex and ChatGPT Work account for more than half of output tokens in its EU sample of ChatGPT Enterprise customers.
Where the growth is strongest
The report does not give one headline number for the whole region. It gives country figures, and they vary a great deal. Reporting by Cryptobriefing and by Cornford and Cross, which both read the report, gives the following figures for the 200-user-or-fewer group and for the share of agent output in the enterprise sample.
| Country | Growth in small-business weekly active users, May to early October 2026 | Agent share of output tokens, September 2026 sample |
|---|---|---|
| Belgium | 12.1 times | Not reported in the sources read |
| Italy | 10.8 times | Not reported in the sources read |
| Luxembourg | 10.0 times | Not reported in the sources read |
| Cyprus | Not reported in the sources read | 81% |
| Austria | Not reported in the sources read | 26% |
| Slovenia | Not reported in the sources read | 22% |
| Lithuania | Not reported in the sources read | 85% |
Every EU country recorded growth of at least five times over the period, according to the same reporting. The agent share figures describe how much of each country's sampled output came from these tools. They are not national adoption rates.
Why it matters for businesses that automate work
For a business that already runs some automation, the report points to a change in the kind of tool being used. A chat window answers one question at a time. An agent is given a goal, works through several steps, and returns a finished result for review. OpenAI itself frames the shift in those terms.
The report also shows a gap in intensity. In September, EU frontier firms, defined as the top 10% of OpenAI's Enterprise and Business customers by output tokens per active user, generated more than four times the output of typical EU firms per active user. Frontier firms outside the EU generated more than twice the output of EU frontier firms. For a business owner, that means the spread between heavy and light users is wide, even inside one market.
What changes in practice for a business that automates work
Most of the practical change is in how work is handed over. Once a tool can carry a task from start to finish, the business has to decide which tasks it may hand over, which systems the tool can open, and who checks the output before it reaches a customer, a supplier or the tax office. A five-person firm can answer those questions with a short written policy and one named reviewer. A group with thousands of staff needs access rules, logs of what the agent did, and a clear owner for each process.
There is also a staff obligation to plan for in the EU. Article 4 of the EU AI Act has applied since 2 February 2025 and requires providers and deployers of AI systems to take measures to ensure AI literacy among staff and others who use AI on their behalf, according to a Traverssmith note on the rule. Some legal commentary says the wording was softened in 2026 so that it asks for measures to support AI literacy rather than a guaranteed level of it. In practice, that means a short training record for anyone who runs agent tools, not a certificate.
Automation also has limits that the report does not address. An agent can produce a wrong figure, misread a document, or act on stale data. It does not decide what a fair price is, what a contract allows, or whether a customer complaint should be escalated. Those decisions still need a person. Businesses should also be cautious about reading the OpenAI figures as a benchmark for their own staffing. The report does not say how many firms are in the sample.
What the report does not show
The numbers come from OpenAI's own customer accounts. The report does not give the size of the EU sample, and it does not show how many businesses use the tools. The token measure counts generated text and code. It does not measure whether a task was finished correctly, how much time was saved, or what the work was worth. OpenAI itself presents the report as a view of its customers, and it links the growth to a call for wider access, deeper skills and more intensive use. Readers should treat the report as evidence of tool uptake, not of economic gain.
Some secondary analyses quote a different headline figure. Cornford and Cross report that OpenAI puts Codex and ChatGPT Work at 55% of output tokens across EU Enterprise and Business accounts by 30 September 2026. That is a broader account cut than the "more than half" enterprise figure in OpenAI's own text, so the two should not be mixed.
What to watch next
- Whether OpenAI publishes the sample size or account counts behind the country figures.
- Whether national statistics offices in the EU publish business surveys that show how many firms use AI agents.
- How the AI literacy duty under Article 4 is applied and enforced by national authorities.
- Whether the country gaps narrow over the next two reporting periods.
How AiStaffo would automate this
Take a business that receives supplier invoices by email and keys them into its accounting package by hand. AiStaffo would connect the mailbox, the document folder and the accounting system, so that an agent reads each invoice, matches it against the purchase order, and posts it when the details agree. Anything that does not match goes into a review queue with the reason written beside it. The owner still approves payments, decides on disputes and sets the rules for what the agent may post. Book a free automation audit.
Questions people ask
What did OpenAI report about EU small businesses and AI agents?
Are the OpenAI figures a survey of all European businesses?
What do the 22% and 85% figures mean?
What is an AI agent compared with a chatbot?
Does the EU AI Act require staff training when we use AI tools?
Should a small business start automating tasks because of this report?
Book a free automation audit
Thirty minutes. We look at one process you run every week and tell you exactly what an AI worker would take off your desk, and what it would not.
