Unleashing the Power of Agentic AI: Five Questions, Five Answers

The rapid development of Artificial Intelligence (AI) is now influencing almost every area of the economy. Companies and organizations are at the beginning of a profound transformation in which AI is being integrated into work processes not just as a supporting tool, but as an intelligent agent in its own right. However, while the opportunities are manifold, there are also challenges, risks and strategic issues that need to be mastered in order to take full advantage of this technology.

In the following, we take a look at current developments together with Dr. Markus Schütten, Senior Manager at MHP. We look at the latest trends and advances in AI, which are associated with challenges such as the integration and company-wide use of AI agents. At the same time, we will shed light on the risks associated with implementing AI in complex corporate structures and show how targeted, strategic planning can ensure the success of these technologies.

Whitepaper: Unleashing the Power of Agentic AI

1) How is AI currently evolving, and what trends are emerging?

AI development and its practical applications are advancing rapidly, leading to modern AI agents that integrate language models (Large Language Models – LLMs) with planning algorithms. These agents have mechanisms that enable them to use tools—referred to as “tools” in agent terminology—to actively interact with their environment. Additionally, they incorporate both short- and long-term memory to retain interaction sequences with users or other AI agents. These advances allow AI agents to autonomously plan workflows based on specific tasks, draw conclusions from environmental feedback, and make goal-oriented decisions.

This also expands their potential fields of application in the future: These now go far beyond isolated, typically very specific tasks and extend to the execution of complex (sub)processes. Similar to human users, the agents can also utilize software tools for this purpose. Possible application scenarios include the autonomous creation, maintenance, and/or editing of customer records in a CRM system, the automated updating of inventory levels in the company’s relevant software solution, or largely automated invoicing in an SAP system.

For very comprehensive and complex (sub)processes or workflows, companies can also use multi-agent systems (MAS), in which they combine multiple AI agents with specialized skills and orchestrate their tasks.

2) What challenges does AI pose for companies?

To derive maximum benefit from the introduction of AI-based agents, companies need, above all, a clear AI strategy and a systematic approach to inventorying their enterprise architecture (EA)—because in the long term, EA provides the framework, and thus the backbone, for any successful work with AI agents.

To date, most companies have primarily used AI in isolated pilot projects or only selectively—often for very specific application areas such as customer management with chatbots or product optimization. However, this fragmented approach limits the transformative potential of AI. While it enables incremental improvements in specific tasks and increases efficiency and productivity there, it does not encompass the company as a whole. As a result, it often fails to focus on the application areas where companies could achieve the greatest added value with AI: particularly, through greater process efficiency and greater innovation capability across all business areas.

What is needed above all are concepts that contribute to the systematic integration of company-wide AI solutions. Chief Information Officers (CIOs) are therefore faced with the critical challenge of developing coherent approaches that help overcome departmental silos and prepare their companies for the orchestrated use of modern AI agents across departments, functions, and divisions. A promising starting point for this is a goal-oriented architecture management (EAM), for example, based on TOGAF (The Open Group Architecture Framework). This enables the seamless, technology-independent integration of AI agents and their capabilities into a company’s architectural development.

In addition, the successful use of AI is accompanied by two other key challenges: It requires active employee participation in the implementation and a solid technical foundation.

3) What risks (and potentials) arise from the use of AI?

Modern AI agents currently perform a wide range of tasks in companies: This ranges from automating repetitive tasks to taking over complex (sub)processes and workflows. They can now even break down unfamiliar tasks into individual steps and then execute them in an orderly manner: For example, a specialized AI agent can create and process customer orders largely independently—and even perform availability checks. If the product requested by the customer is not available, it will identify and report this.

Thanks to these developments, AI agents can now support employees even with more demanding tasks. This frees them up to focus on value-added activities. Modern AI applications therefore offer tremendous potential for companies, but their implementation also comes with risks:

• Economic Risks:
  Many companies currently rely on a single provider to operate their AI agents and systems. This can become problematic if the provider changes its strategy, increases prices, or even disappears from the market. Existing users would then be forced to undertake complex and expensive migration projects to transition their AI systems to a new platform.

  To minimize this risk, companies should adopt a vendor- and model-agnostic AI approach. The enterprise architecture should be designed to facilitate the deployment and integration of new AI models without any technical complications. This reduces reliance on individual vendors and ensures that users can continuously benefit from the latest developments and innovations in AI.

• Technical Risks:
  From a technical perspective, the greatest risk arises from security. If a company’s IT department doesn’t set boundaries for an AI agent—for example, by clearly defining which data it can use and which tools it can access—employees can potentially access business-relevant information through “smart” instructions that were not intended for their role and tasks.

  The IT department can significantly minimize this risk by assigning clear permissions and scope of action to each agent: These are predefined instructions or directives that establish the basic context of the AI agent and thus determine its behavior in all interactions. 

4) How can the optimal (strategic) use of AI be derived from the enterprise architecture?

A promising approach is working with enterprise architecture reference models, such as TOGAF (The Open Group Architecture Framework). The exact procedure is very complex, so we discuss it in detail in our paper: “Unleashing the Power of Agentic AI”. Below, we present it in simplified form in three steps:

• The Business Capability Map as a Foundation
  EA models typically use business capability maps to provide a structured representation of the company’s capabilities. Planners can then break down each capability in this map into individual components such as resources, processes, information, and users. This clearly defines the environment in which an AI agent deployed for each of the company’s capabilities would need to interact.
• Identifying Possible AI Application Scenarios
  By switching to the process perspective, planners can then identify the most valuable AI application scenarios. To do this, they select those processes or workflows that are particularly easy to automate or that incur high manual processing costs, thereby justifying higher investment volumes.
• Assigning the Resources and Information Required by AI Agents for Process
  Finally, the planners assign the future AI agent the resources and information it needs to execute its assigned process. They also need to define which data it can access and which tools it can interact with. 

It is also possible, in principle, to link several agents to form multi-agent systems (MAS). These can sometimes even handle entire end-to-end processes. Planning an MAS follows a similar process model, but is significantly more complex—after all, the company must also specify how the coordinating instances should work. There are various technical options for this. You can find further details on this in the paper: “Unleashing the Power of Agentic AI”

5) In summary: What do I need to consider as a company in order to use AI agents successfully and profitably?

The following criteria are crucial for the successful use of AI agents:

• A Strategic Approach
  Companies can optimally benefit from the advantages of AI agents if they deploy them not sporadically but strategically—where they generate the most value for the business (see point 4 for more details).

• A Solid Technical Foundation
  More complex tasks are often handled not by individual AI agents, but by “multi-agent systems” (MAS). In these systems, several agents work closely together. Each agent is responsible for a specific skill or process segment, thus contributing to the overall process. This requires orchestration instances to coordinate the work of the individual agents.

  To utilize such instances, however, companies require an “AI operating system.” This provides the technical foundation—across manufacturers—for all AI agents and multi-agent systems used there. One example is IntelliCore from MHP: This solution can integrate AI-based systems and connect them to various machines and software tools via APIs.

• Early and Active Employee Involvement
  Advanced AI agents typically rely on sophisticated language models and are therefore able to understand and execute even complex instructions in natural language. Nevertheless, AI agents remain tools. Accordingly, the quality of their results depends heavily on whether and to what extent companies involve their employees in AI-based processes in the future, enabling them to intervene if necessary. We therefore generally recommend continuing to focus on employees and pursuing “Human-in-the-Loop” approaches: AI agents engage with users in an iterative process in which they work together, step by step, to achieve the optimal result.