EU small businesses now use AI agents eight times more. What that tells you.

On 14 October, OpenAI published a report on how European companies use its tools.¹ The headline number is for small businesses. Since May 2026, weekly active users of Codex and ChatGPT Work among small and medium EU businesses grew more than eightfold, with growth in every member state.²

OpenAI counts a small or medium business as one with 200 or fewer associated users.³ An agent, in its words, is a tool that carries out a task from planning through completion, instead of answering one question at a time.⁴

Here is the part that matters. Eightfold growth tells you the number of people grew. It doesn't tell you how many firms, or how much money they save. OpenAI gives no base figure in the post, so we can't say whether eight times a small number is a lot.⁵

The country gap is easier to read. Among ChatGPT Enterprise accounts in OpenAI's sample, the share of output tokens (the words and code the model writes) that came from Codex and ChatGPT Work ran from 22% in Slovenia to 85% in Lithuania.^6,14 Across the EU sample, those two tools now make up more than half of output.⁷

OpenAI defines frontier firms as the top 10% by output tokens per active user.⁸ In September, EU frontier firms generated more than four times as many output tokens per user as typical EU firms.⁹ Frontier firms outside the EU generated more than twice as many as EU frontier firms.¹⁰ EU firms are 10% of frontier firms worldwide.¹¹

Two limits apply. This is one vendor's data, covering only its own Enterprise and Business accounts.¹² A firm that uses another vendor, or none, isn't in it.

If you run a small firm, the useful signal is not the eightfold figure. It is that agent tools are now ordinary enough to be in use in every member state. Pick one repeated task, such as sorting supplier invoices or drafting the monthly report, and run it through an agent for a week. Check the output by hand each time.

If you build software, expect customers to ask what your product does when an agent, not a person, is the one clicking. Check that your data export and your forms work without a human at the screen.

OpenAI's own conclusion is that the gap to close is depth of use, not first use.¹³ We'd add that depth only pays when someone checks the work.