Why AI agents pay off – and how to keep them under control

Companies in Germany have high expectations for agentic AI: 90% of them believe AI agents have the potential to fundamentally transform their organizations. The problem is that only 4% of companies feel fully prepared for the productive use of AI. These figures were revealed in the latest SAP survey, “The Value of AI 2026.” A key finding is that the risks associated with agentic AI are evolving much faster than the governance needed to steer AI use along safe and orderly paths. Furthermore, in many companies, the business value of AI investments cannot yet be clearly measured.

Key takeaways:

• The SAP survey “The Value of AI 2026” shows that companies are investing heavily in artificial intelligence (AI).

• These investments are yielding satisfactory results and added value for companies.

• At the same time, there is a lack of transparency and control regarding the AI agents used within the company.

• Employee enablement is a critical prerequisite for AI success.

Governance lags behind AI

The existing governance gap is supported by the figures of the survey: Only 11% of companies consider their ability to effectively manage AI agents to be sufficient. More than half of those surveyed (57%) have no processes in place to monitor AI agents through human oversight. Transparent documentation of AI agents’ authorizations, which would allow companies to track which agent is authorized to access what, is currently available in just under one-third of companies.

Employees have long been using their own AI agents without the IT department’s knowledge or ability to control them. It is particularly critical when users copy business content into their personal accounts with popular AI assistants and large language models (LLMs) – bypassing any central control. As a result, hundreds or even thousands of so-called “shadow agents” – AI agents operating within the company without central registration, an effective authorization concept, or proper oversight – emerge unnoticed.

What structured governance means in practice

Governance and transparency are not just compliance exercises, but essential prerequisites for the economic success of AI investments. Accordingly, governance is one of the four most important criteria for companies when evaluating the benefits of an AI project. The others are expected business value, technical feasibility, and the availability of the necessary data.

When granting authorizations to AI agents, companies should model their approach on the lifecycle of human identities – including onboarding and offboarding. Each agent requires a defined registration process, clearly assigned permissions, and regulated access to data sources. Likewise, employees must be able to track what an agent is doing in the system at all times and verify its results. Continuous monitoring ensures that the agent consistently operates within the specified parameters and actually delivers the expected business value.

Positive experiences drive AI adoption

47% of German companies are satisfied with the return on investment (ROI) of their AI investments. At the same time, 77% say they are still far from realizing AI’s full potential. These figures illustrate a learning curve that is typical of Agentic AI’s current level of maturity: Companies initially invest in individual, clearly defined AI use cases and are satisfied with the added value achieved there. These positive findings heighten their awareness that there are many more potential scenarios for AI within the company. The more experience a company gains with AI, the greater its awareness becomes of the additional potential that can still be tapped.

Factors that limit the potential of AI

For this potential to be fully realized, two things are needed: a comprehensive overview of the AI agents and use cases deployed within the company, and the AI enablement of employees. This is because insufficient employee training, along with poor data quality, ranks among the most common obstacles to successful AI deployment.

Without employee enablement, AI adoption will fall short of its potential – regardless of how powerful the technology being used is. Users must understand how generative AI works in principle and where its limitations lie in order to meaningfully interpret and evaluate the results produced by AI agents.

Why semantics and business context are crucial

The deep integration of AI agents into a company’s own systems and the comprehensive process knowledge are intended to give SAP a strategic advantage over other providers in the realm of enterprise AI. SAP Joule makes the semantic relations between tables, business objects, and data fields usable for AI applications, rather than simply allowing agents to access isolated data points. This context is crucial for precise answers: AI agents need access to consistent, context-rich enterprise data to reliably solve complex optimization tasks. It is precisely where this context is available that AI agents deliver the greatest time and efficiency gains.

The added value of AI today and in the future

Currently, AI delivers its greatest added value in three areas: decision-making, customer interaction, and gaining new insights. The benefits stem primarily from the fact that information from structured and unstructured data sources is consolidated and made accessible via natural language. This enables companies to gain insights that were previously hidden.

Increased productivity, on the other hand, currently plays a relatively minor role as a value-added factor. Looking ahead, however, it will become the primary motivation for AI adoption as soon as companies are able to consistently translate these new insights into more efficient business processes. The shift from mere insight generation to actual process efficiency is thus the next stage of maturity that companies will reach through the use of AI.

Conclusion: How companies are harnessing the transformative power of AI

The survey “The Value of AI 2026” makes it clear: Investments in AI are already paying off for companies today, and agentic AI is perceived as a disruptive, transformative force. Economic success hinges on the ability to centrally manage AI agents rather than allowing them to proliferate unchecked within the company. Those who invest today in authorization management, traceability, and employee enablement are laying the foundation for turning individual AI successes into sustainable, company-wide value.