AI agents: high adoption, uncertain governance

The question is no longer whether companies will use AI agents, but who will govern them. That is what the "SAP LeanIX Agentic AI Survey 2026" shows, a survey carried out in March 2026 among 228 enterprise architecture (EA) leaders, 70% of them at European companies.

Key findings

• Widespread adoption. 75% of companies use AI and 44% already deploy agents, almost always in limited functional areas. Another 40% are experimenting and 14% plan to: altogether, 98% are on the way.
• Little visibility. More than 80% do not know how agents support business capabilities, their performance, their compliance or their return on investment. In addition, 38% do not even intend to achieve that visibility.
• Main obstacle: people. 58% point to employees' lack of knowledge and the difficulty of identifying use cases; technology limitations weigh on only 15%.
• Governance under construction. Close to three quarters consider specific rules for agents necessary: 22% already have a framework and 51% are developing one. In 72% of cases, an agent requires prior review.
• Blurred responsibilities. Only 37% have a designated area to manage agents; in 48%, several teams are involved with no clear owner.

The enterprise architects' gap

Architects take part in managing agents in barely a third of organizations and lead the topic in only 5% of the cases with a designated area, almost always IT. Yet 63% believe their discipline should drive the adoption and governance of AI. The study suggests a reason: they are comfortable with the inventory of agents and their relationship to business capabilities, but few take on the full lifecycle; only 27% believe it falls to them to lead performance tracking.

Conclusion

Governing agents requires clear responsibilities, visibility into their value and risk, and performance metrics. Those who solve this first will turn experimentation into measurable results. A good starting point is to inventory the agents, designate an owner and define how their return will be measured.

Source: SAP LeanIX, "Agentic AI Survey 2026" (n = 228, March 2026). Summary and own commentary.