OpenAI says agent tools generate over half of tokens in its EU enterprise sample

OpenAI sees AI agents as a route to stronger European businesses. Its research published October 14 shows a shift in usage, rather than a measured economic payoff: Codex and ChatGPT Work generated more than half of output tokens in its EU sample of ChatGPT Enterprise customers. That measures generated output, not the share of tasks successfully completed.

The report, from OpenAI Economic Research, examines businesses’ move from asking AI questions to having agents carry out tasks from planning through completion. It groups Codex and ChatGPT Work as agentic tools—software intended to do work for people, rather than only answer them. The headline finding concerns those two products’ share of output within the enterprise sample, not the percentage of European businesses using agents.

Rapid growth, uneven use

The shift also reaches smaller customers. Weekly active users of Codex and ChatGPT Work among EU small and medium businesses grew more than eightfold since May 2026, OpenAI says. Growth occurred in every member state. That is a different measure from the enterprise finding: it tracks people using the tools each week, rather than how much output those tools generate.

Growth across the bloc does not mean agent use looks the same everywhere. Among sampled ChatGPT Enterprise accounts, the portion of output generated through the two tools varied sharply by country. Slovenia and Lithuania marked the low and high ends of that comparison.

Agent tools’ share of enterprise output

Codex and ChatGPT Work% of output tokens

22%85%

OpenAI’s sampled ChatGPT Enterprise accounts show a country-level range, not a measure of adoption among all businesses in either country.

The frontier is defined by output

A separate comparison in the same report asks how intensively firms use AI. Here, OpenAI’s frontier-firm analysis includes only ChatGPT Enterprise and Business accounts. Its definition of a leading firm is based on output per active user—not revenue growth, profitability or a measured improvement in the quality of work.

Two definitions behind the findings

• Small and medium businesses: businesses with 200 or fewer associated users. This is the group used for the weekly-user growth finding.
• Frontier firms: the top 10% of firms by output tokens per active user. This is the group used for the usage-intensity comparison.

In September, EU frontier firms generated more than four times as many output tokens per active user as typical EU firms. Yet frontier firms outside the EU generated more than twice as many as the EU frontier group. The comparison therefore identifies two gaps: one within the EU customer sample and another between its most intensive users and their counterparts elsewhere.

EU-based firms also made up 10% of frontier firms globally in the analysis. That figure describes where OpenAI’s highest-intensity business customers are located. It should not be read as Europe’s share of the world’s most productive companies: entry into this group depends on the report’s usage measure and its Enterprise and Business account sample.

Economic gains remain the ambition

OpenAI’s proposed response is to help businesses move beyond experimentation, build workers’ skills and confidence, and make advanced tools more accessible and affordable. CFO Sarah Friar ties that agenda to stronger businesses and growing economies. Those are the benefits the company wants deeper adoption to deliver. They are distinct from the report’s findings on user growth, generated output and differences in usage intensity.