1. July 2025 By Attila Boka
How do you actually use Agentic AI?
This article is the second part of our three-part series of specialist articles on the topic of "Agentic AI". The aim is to provide a solid foundational understanding of agent-based AI systems as a basis for making informed decisions on future investments and organizational developments. Instead of technical details, the focus is on clear, relevant strategic questions that inform business decisions.
In this series, the following articles will be published in succession:
1. What is Agentic AI and how does it work?
2. How do you actually use Agentic AI? (this article)
3. What are the use cases for Agentic AI?
From idea to implementation - how companies can introduce Agentic AI
The first article in this series explained the conceptual foundations of Agentic AI. This article addresses the next step: the journey from initial idea to practical implementation within a company. It focuses on the specific prerequisites, challenges and design areas that companies are currently navigating as they plan and prepare their first Agentic AI initiatives. This includes not only technological requirements, but also organizational alignment, strategic integration with existing platforms, and the establishment of new roles and responsibilities. The goal is to provide a sound basis for action that enables decision-makers to harness the potential of Agentic AI in a focused, scalable and business-oriented manner.
Three central introductory questions: What? Where? With what?
Before introducing Agentic AI, it’s essential to clarify the initial context, overarching goals and operational levers. Companies should view Agentic AI not just as a technology project, but as part of a broader transformation initiative that affects processes, roles, system landscapes, and decision-making logic. Defining a strategic starting point forms the basis for a viable target picture - whether aiming for an initial minimum viable product (MVP) or a broader rollout. Three key questions help structure and realistically assess the project:
- What is the goal? Is it operational relief, quality improvement, new customer experiences, or long-term innovation? The strategic objective sets the direction and helps prioritize both initial MVPs and future scaling. Decision-makers should be able to clearly articulate and communicate the concrete business value - to both budget managers and implementation partners.
- Where is the greatest leverage? Agents are particularly effective in areas with repetitive, data-driven, decision-intensive, or non-deterministic processes. They add value especially in interfaces with high interaction (e.g. between customer channels, logistics systems or specialist departments). These processes are often closely tied to KPIs such as throughput time, quality or customer satisfaction.
- Where do we begin? Successful projects often start “in the shadows” - in areas that don’t impact critical core processes but can still deliver visible impact. Examples include internal planning, reporting procedures, or structured communication routines. These entry points allow teams to gain experience, establish technical foundations, and introduce the organization to new work practices - without high entry barriers or strategic risks.
How agents create concrete added value - initial target areas
Agentic AI is not an end in itself. It generates value where existing systems reach their limits, for instance, in decision-making under uncertainty or when rule-based processes fall short. Agents can stabilize, control, and adapt-particularly where decision logic affects multiple departments or is constantly evolving due to changing conditions. They act as an adaptive control layer in processes that are too complex for full automation or manual management, such as prioritizing tasks, allocating resources dynamically, or supporting day-to-day decisions with data-driven recommendations.
Typical target areas:
- Customer service: Agents handle repetitive inquiries completely and prepare complex cases at the same time, including bundling information, analyzing sentiment, and suggesting escalation paths. This boosts service quality while reducing costs.
- Procurement & Logistics: Agents act as early warning systems for supply issues, recommend alternative routes or suppliers, and can place orders under defined conditions.
- Sales & Marketing: Sales agents identify leads in CRM systems, assess interactions, and suggest personalized outreach strategies - based on data and continuous learning.
- Product development: Agents analyze market feedback, assess product usage, and help prioritize development features.
- Back office and compliance: Agents assist with tasks like invoice verification, contract validation, and compliance reporting, easing the burden on experts by pre-processing data and offering structured decision proposals.
- Workforce management: Agents support shift scheduling, capacity forecasting, and skill allocation, especially in dynamic environments with fluctuating demand or limited resources.
Example of experience: Ongoing initiatives are exploring the use of agents in demand planning, such as analyzing order trends, seasonal fluctuations, or delivery times. Early findings indicate that this creates adaptive, data-driven procurement processes that are being\ developed iteratively in collaboration with specialist departments. The emphasis is on continuous learning, targeted automation, and the evolution of robust decision-making logic.
Internal requirements: data, systems, governance
Creating agent-ready structures requires not only technical infrastructure but also clear definitions of data access, decision-making logic, security standards, and responsibilities. A company’s ability to successfully deploy agents depends largely on how well its data, systems and governance mechanisms are aligned.
- Data situation: Data availability, quality and freshness are critical success factors. Agents must be able to access relevant information in real time - both structured (e.g. product data, master data) and unstructured (e.g. emails, tickets). In addition, a minimum level of semantic tagging is needed for agents to interpret context.
- System landscape: A flexible, API-based IT architecture is a major advantage. Legacy systems with poor interfaces or outdated documentation can hinder agent performance or make their use cost-ineffective. Not only APIs are important here, but also a clear understanding of how data flows can be orchestrated and controlled.
- Governance: Clearly defined responsibilities for agent development, monitoring, and outcome validation. Transparent control mechanisms are essential to maintain trust and prevent misuse. Early consideration should be given to auditability, version control and decision.
- Access and rights concepts: Agents require controlled system and data access within strict security protocols. Determine whether agents should function as technical users or require dedicated identity and access management.
- Monitoring and checkpoints: To build trust and minimize risks, checkpoints must be built in - for critical action approval,escalation in the event of uncertainties or for logging decisions. These controls are key for scaling and organizational acceptance.
Observation from customer initiatives: In heterogeneous IT landscapes with complex system responsibility, many companies currently see great potential in building agent-capable operating models. This capability is being discussed as a strategic lever, particularly with regard to future scaling, governance specifications and integration requirements, even if concrete empirical values are still being developed in many cases.
Team dynamics and change: interaction between people and agents
The introduction of Agentic AI shifts tasks, roles and expectations gradually but noticeably. Employees begin taking on new responsibilities while agents handle routine work. This creates new points of interaction between people and intelligent systems. Human roles evolving into guiding, instructing, and supervising AI. Transparency, trust, and clarity become essential in both daily work and process oversight. It is crucial to actively shape this change through clear role allocations, targeted enablement measures and a culture of learning.
Important aspects:
- Build acceptance: Employees need to understand how agents function, their limitations,and how roles are changing. Without communication, skepticism or resistance may grow
- Create new roles: In addition to traditional roles such as project management or data analysis, new functions such as agent designer, agent supervisor or agent product owner are emerging. These combine technical understanding with business insight.
- Strengthen cross-functional collaboration: Developing agents require close cooperation among IT, specialist departments and data teams. Silo thinking prevents viable outcomes
- Enablement and training: For agents to be used effectively in everyday life, targeted enablement measures are needed for specialist users - for example in training, guidelines and interactive instructions.
- Establish feedback loops: Continuous feedback from users to the agent system and its support team increases transparency, improves agent logic and drives organization-wide acceptance.
- Moderating the willingness to change: Agentic AI changes task profiles and work processes. Active change management supports cultural anchoring and reduces resistance to automation initiatives.
Agentic AI is also a catalyst for organizational learning and a cross-divisional culture of innovation. By making existing processes visible, systematically recording decision-making logics and integrating new context dynamically, agents enhance a company’s ability to reflect, adapt, and innovate. This fosters collaboration across IT, business, and governance, as well as driving operational excellence and new digital opportunities.
Find out more in our webinar
We will cover these and other topics in more detail in our Agentic AI webinar, which will take place in fall 2025. You can find more information about the webinar here and register directly to take part live.
Realistic planning: piloting, scaling, anchoring
Agentic AI is like other large-scale innovations, except for one decisive feature: it impacts technology, processes, and culture simultaneously. Traditional transformation principles apply accordingly. Instead of relying on disruptive change, an iterative approach with a pilot character, realistic target images and a clear scaling perspective is recommended. Stages defined at an early stage help to limit risks, build up experience and strengthen the organizational capacity for change in the long term.
Recommended steps:
- Piloting: Start with a use case that poses minimal risk, offers high visibility, and has existing data access. The primary goal should be to test functional feasibility and gather feedback from the specialist department. These initial agents should be designed to create specific learning opportunities for the organization, both technically and operationally. Stakeholder input should be systematically collected to identify potential barriers to adoption or weaknesses in the design early on.
- Scaling: Expand the agent model to similar processes or adjacent organizational units. It’s important that the system architecture and governance structures evolve in tandem. Consider creating an internal agent portfolio early on - complete with standardized evaluation criteria, resource planning, and a central accountability structure (e.g. an agent owner or operations manager).
- Anchoring: Establish a standardized framework for agent operations, including an operating model, performance indicators, access rights management and development cycles. This also involves implementing monitoring and audit processes, along with rules for version updates, escalation, and training. Leading organizations use this phase to embed their Agentic AI strategy into the broader context of digital value creation - aligned with corporate goals, regulatory requirements, and cultural transformation.
The objective: build sustainable, organization-wide capabilities to develop, manage and operationalize agent-based automation. This includes clearly defined responsibilities, anchored governance, and the flexibility to integrate new agent roles into everyday business. Agentic AI should not be treated as a standalone solution, but as a strategic competency embedded across the organization.
Leveraging existing investments: Conversational AI as a bridge
Many companies already have experience with conversational interfaces such as chatbots or Copilot. These solutions offer a strong starting point for Agentic AI because they already incorporate essential components like context recognition, dialog management, and automated response generation. Transitioning from a reactive bot to a proactive, goal-driven agent is technologically challenging but conceptually feasible, especially if APIs, usage data, and access management structures are already in place.
The value lies not only in reusing technical components, but also capitalizing on existing organizational learning: Teams have become familiar with working alongside AI-based systems, articulating functional requirements and understanding limitations. Agentic AI can build on this by adding autonomous decision-making, goal tracking, and continuous learning to existing capabilities. In doing so, a "talking interface" evolves into an operational agent that enhances and expands the strategic utility of dialog systems.
- Leveraging existing infrastructure: APIs, intent models, and logging systems from chatbot solutions can often be adapted into agent-based architectures. Workflow engines and data pipelines already in use can serve as a technical foundation for agents.
- User acceptance is already there: Customers and internal users are accustomed to conversational interfaces which facilitates the introduction of new agent-based functions. This familiarization not only reduces the training effort, but also lowers the barrier for adopting more advanced, agent-powered features.
- Using data for further development: Chatbot interaction data offers valuable training materials for agents, for example, to improve task prioritization, classification or escalation behaviors. This data also supports performance optimization through feedback loops.
- Step-by-step expansion possible: Companies can iteratively transform existing bot setups into agent-like systems, starting with simple decision logic, and later introducing goal-driven autonomy and adaptive behaviors.
- Front-end remains stable: The user interface remains familiar to end users, even as more complex agent functionality is added in the background. This simplifies adoption and boosts acceptance.
- Making organizational learning usable: Existing chatbot initiatives have often already established roles, support structures, and technology standards internally, providing a ready-made foundation for extending into agent-based systems.
"Lessons learned" from customer projects
Organizations that didn’t treat conversational AI as a stand-alone communication tool- but instead integrated it into a broader architecture and governance model- were better positioned to adopt Agentic AI successfully. This foundational work made it easier to later deploy more sophisticated agent roles in a data-driven and secure manner. In such cases, conversational AI has been reframed not just as an interface, but as an adaptive entry point into agent-driven business systems. Teams that incorporated structured user feedback, monitoring mechanisms, and system integration early on saw faster and more successful transitions into Agentic AI - technically, organizationally and strategically.
Practical tips
A simple evaluation matrix (based on data availability, process maturity, regulatory pressure, and business impact) can help to identify suitable pilot processes. It can help to prioritize strategically viable projects, manage expectations towards stakeholders and prepare coordination with architecture or budget managers in a more targeted manner. Particularly in the early planning phase, it enables a structured discussion about which use case is both technically feasible and organizationally compatible - an important step to avoid friction losses at a later time.
For strategic decision-makers: Assess early how well Agentic AI aligns with your existing platform architecture and IT governance policies. Financial implications, such as efficiency gains through automation or reduction in process costs, are especially important for CFOs. Metrics related to scalability, control and impact on KPIs like time-to-market, OPEX, or customer satisfaction should also be considered. For CEOs and CDOs, it’s important to understand how Agentic AI contributes to broader strategic goals such as innovation speed, market differentiation, or resource optimization.
For specialist departments: Evaluate which decisions agents can support, and how much control or transparency is needed. Clearly define which tasks can be delegated, how feedback mechanisms should work, and where final decision-making authority remains with human users. Business leads should also analyze how agent-based processes can affect customer experience, service quality, or time-to-market. Additionally, consider how Agentic AI will shift or enhance role profiles and what training, change management, or operational support will be required as a result.
Conclusion and outlook: from experiment to strategic component
Agentic AI is rapidly evolving from a frontier technology into a business-critical capability. Organizations that proactively build the right infrastructure - both technical and organizational - will be best positioned to capitalize on this transformation, improving performance while strengthening long-term competitiveness.
Three final recommendations:
- Clarify your organization’s maturity in terms of Agentic AI - technologically, organizationally and culturally.
- Start with a viable pilot project that demonstrates visible added value and builds internal trust
- Think in terms of structures: governance, reusability and scalability must be considered from the outset.