Europe Crossed the AI Agent Threshold. The Hard Part Comes Next.

On October 14, 2026, OpenAI's economic research team published a ten-page report with a thesis that enterprise software vendors have been pitching for years, and which the data finally appears to support: European workers are no longer just asking AI questions. They are handing it tasks. The report, titled "Europe's Agentic Opportunity," draws on usage data from ChatGPT Enterprise and Business accounts across the European Union, and its headline number is striking. By September 30, agentic tools accounted for 55 percent of all output tokens generated by EU business customers.

The breakdown inside that number matters. Codex, OpenAI's coding agent, generated 34 percent of output tokens; ChatGPT Work, the task-oriented agent launched in July, generated 21 percent. Ordinary conversation made up the remaining 45 percent. Four months earlier, in May, the agentic share stood at just 21 percent. The figures use a 28-day trailing window, so this is not a one-day spike. It is a sustained pattern of work.

From questions to delegation

The conceptual shift is the real story, not the statistics. A chatbot answers a prompt and the session ends; the human does the rest. An agent takes a task from planning to completion, drafting the document, running the code, booking the follow-up, chaining tools together across a workday. OpenAI frames this as Europe's new era of AI: CFO Sarah Friar said the continent is entering an era "with tools that can carry out tasks for people." The report argues that businesses must move beyond experimentation and put agents to work on real processes, pointing to better access to advanced AI capabilities, plus workers' skills and confidence, as the decisive factors.

There is reason to take the framing with a measure of skepticism. The data describes OpenAI's own products, measured by OpenAI's own research team, using OpenAI's own metric of output tokens. Token volume is a measure of activity, not of value: a token-heavy agent can be busy without being useful. But the direction of travel is hard to dismiss. Weekly active agent users at EU small and medium businesses grew 8.1 times between early May and early October, with every member state recording growth above five times. When hundreds of thousands of workers change how they spend their day, something structural is happening.

Token volume is a measure of activity, not of value. But the direction of travel is hard to dismiss.

The small-business surge

The most surprising finding may be where the growth is fastest. Agentic AI was supposed to be an enterprise game, sold top-down through IT departments and pilot programs. Instead, small and medium businesses are moving first. Weekly active agent users at EU firms with 200 users or fewer grew 8.1-fold since early May, and most of that growth arrived after ChatGPT Work launched in July. Belgium led at 12.1 times, followed by Italy at 10.8 and Luxembourg at 10.

That matters economically because small firms are the EU's economic backbone. The report notes that SMBs account for 63 percent of private-sector employment in the bloc and contribute more than 5 trillion euros annually. Agents reach these companies differently than previous waves of enterprise software did: no implementation consultants, no six-month rollouts, no procurement committees. A credit card and a browser are enough. The diffusion pattern looks less like traditional software adoption and more like the smartphone: bottom-up, fast, and uneven.

A map of two Europes

Adoption is anything but uniform. Lithuania led the bloc with agentic tools producing 85 percent of business output tokens in September, followed by Cyprus at 81 percent, Hungary at 79 percent, Poland at 76 percent and Sweden at 73 percent. At the other end, Slovenia sat at 22 percent, Austria at 26 percent, Portugal at 29 percent and Spain at 30 percent. France landed at 55 percent; Germany, the EU's largest economy, at 51 percent, ranking 17th among the 27 member states.

The report notes that smaller countries such as Cyprus, Estonia and Lithuania show relatively high agentic use within their larger firms, which hints at why the pattern forms: tight, export-oriented economies with digital-first workforces adopt faster than larger markets with legacy IT estates. Germany's middling position is the cautionary tale. Economic weight does not automatically translate into technological depth, and the firms with the most to gain may be the slowest to move.

The depth gap

Breadth is not depth, and this is where the report turns critical of its own continent. OpenAI defines "frontier firms" as the top 10 percent of businesses by output tokens per active user. EU frontier firms generated more than four times the tokens per user of typical EU firms in September, but frontier firms outside the EU generated more than twice the output per user of the EU frontier group. Europe makes up just 10 percent of global frontier firms; North America accounts for around 65 percent.

In other words, Europe's best AI users are far ahead of Europe's average users, and far behind America's best. The gap between the EU's most AI-active firms and the global frontier is, by the report's framing, an opportunity: adoption has broadened, but intensive use, the kind that rewires processes rather than assisting them, is still rare. Whether closing that gap requires new tools, new skills, or new regulation is the open question. The EU spent the last two years building the AI Act around chatbot-era risks; agentic tools that plan and execute on their own pose different questions, and Europe's rulebook was not written for them.

What it means for work

For workers, the practical takeaway is simpler than the policy debate. The unit of AI work is no longer the prompt. It is the delegated task: defined well, handed over, and reviewed when done. European firms that learn that workflow now, while adoption is still shallow, will define what the frontier looks like when the continent catches up.