AI Agents Booming in Europe: The Depth Gap Decides
In short
The new OpenAI report shows agents are booming in Europe, but the real gap is depth. Typical firms barely move while the top 10% produce four times more output per user. For Swiss SMEs, this means delegating one process end-to-end, not distributing licenses.
The Numbers: Agents Dominate Output, but Depth Is Missing
The report 'Europe's Agentic Opportunity' from OpenAI Economic Research (14 October 2026) shows that by the end of September 2026, agent tools generated 55 percent of output tokens among European business customers. Weekly active usage among small companies grew 8.1× since May. Germany sits at 51 percent agent share, only mid-table (rank 17 of 27). The real story is not the breadth of adoption but the depth of usage.
55 %
Share of output tokens from agent tools among European business customers (September 2026)
The key finding is structural: The top 10 percent of firms (frontier firms) produce more than 4× as many output tokens per active user as typical firms. Measured from May, typical EU firms grew only 1.5×, EU frontier firms 6.1×, and frontier firms outside the EU 12.9×. This is a depth gap, not a breadth gap.
Depth over Breadth: Delegate One Process End-to-End
The implication for your company: Do not simply distribute licenses to as many employees as possible. Choose a single process with a clearly measurable outcome and delegate it fully to an agent. This requires clean interface definitions, integration with existing systems like Abacus, Bexio, or Microsoft 365, and governance that respects GDPR and the EU AI Act. The Schweizer KI-Podcast discusses this approach under the title AI as a leadership topic.
Many Swiss SMEs fail not because of technology but because of poor process selection and inconsistent execution. A common mistake is starting too broadly and drowning in complexity. Instead, automating a single process well delivers more value than spreading automation thinly across many processes. For more strategic guidance, see KI strategy for Swiss SMEs.
A Practical Path: From Licenses to Depth
- Identify a process with high manual effort and a clear, measurable outcome.
- Clarify data access and integration: Abacus, Bexio, Microsoft 365, or other core systems.
- Define rules for the agent, including human approval for critical steps.
- Set a metric: not tokens, but output quality, turnaround time, or hours saved.
- Start small, iterate weekly, and only then expand to the next process.
- Focus
- Breadth: Distribute licenses to many employees. Depth: Delegate one process end-to-end.
- Usage intensity
- Breadth: occasional, often without a clear goal. Depth: daily, end-to-end, with a defined outcome.
- Measurable effect
- Breadth: unclear, many unused accounts. Depth: time savings and quality gains directly visible.
- Governance effort
- Breadth: low, but risk of sprawl. Depth: initially higher, but controllable.
- Typical growth per report
- Breadth: typical firms grow only 1.5×. Depth: frontier firms grow 6.1× (EU) or 12.9× (outside).
Important caveat
The data comes exclusively from OpenAI's own enterprise and business accounts. They measure neither productivity nor hours saved. Token share is not a share of tasks or people. Coding agents produce long files, and individual power users can statistically shift a country. The robust finding remains: The gap opens at the top.
Conclusion: The Depth Gap Is Your Opportunity
The report makes visible what many decision-makers already sense: Merely introducing AI tools yields little if usage remains shallow. Your opportunity lies in thinking one process through to completion and delegating it to an agent. This works without an internal AI team: an external partner can handle selection, integration, and governance. The first step is small: name a process that currently costs too much manual effort.
Frequently asked questions
- What does the depth gap mean concretely?
- Typical firms barely grow in agent usage (factor 1.5 since May), while the top 10% (frontier firms) produce more than 4× as much output per user. The gap arises from intensive, consistent delegation of individual processes, not from broad license distribution.
- Why is Germany only in the middle?
- Germany has an agent share of 51% and ranks 17th out of 27 EU countries. This suggests broad but shallow usage: many accounts, little depth per user.
- How do I start with an AI agent as an SME?
- Choose a process with high manual effort and a clear outcome. Clarify data access and integration (e.g. Abacus, Bexio, Microsoft 365). Define rules and human approvals. Set a metric and iterate weekly.
- What role does GDPR play with AI agents?
- If personal data is processed, GDPR applies. You must ensure the agent uses data only for its intended purpose, access is controlled, and any data processing agreements are contractually regulated. For EU relevance, the EU AI Act also matters.
- Can I introduce AI agents without an internal AI team?
- Yes. An external partner can handle process selection, integration, governance, and operations. This is often faster and less risky than building an internal department.
Sources
Chris Jon Graf · AI Strategist & CEO
For over 15 years, Chris Jon Graf has built productive Swiss software platforms and advised executive teams on the strategic deployment of artificial intelligence. Would you like to evaluate this topic for your company?