Agent death or game changer for the German industry?

According to Gartner, over 40 percent of all agentic AI projects will be discontinued by 2027 - due to exploding costs, unclear business benefits and a lack of risk control. But what about the other 60 percent? Amara's Law may provide an answer: new technologies are overestimated in the short term and underestimated in the long term. This is also the case with Agentic AI - the current impact is limited, but in the long term it will fundamentally transform workflows, decision-making processes and operational efficiency. The key question is therefore: what do manufacturing SMEs need to consider so that AI agents learn to work in their own company?

Between visionary PowerPoint slides and real added value

Agentic AI is often mystified because it is wrongly equated with GenAI. But Agentic AI goes further - these systems not only assist, but also act autonomously, make decisions, control processes and implement goals independently, which creates both opportunities and challenges. In an industrial context, Agentic AI could, for example, autonomously control warehouse processes, dynamically adapt supply chains or proactively avoid production errors - and thus raise these areas to a new level.

Agentic systems promise efficiency gains, faster response times and a higher degree of automation. However, their introduction requires more than just technical feasibility. It requires a deep understanding of processes, responsibilities and economic objectives. Medium-sized industrial companies in particular are (still) very reluctant to do so. This is understandable, as many companies have not even strategically integrated GenAI yet. For real change to happen, AI needs to go where it really makes a difference: in the specialist departments, and away from 'me-trying-it-out mode' and into systematic practice.

AI needs to get out of the chair circle and into an overall strategic concept

This applies even more to Agentic AI. Premature projects without a focus on benefits, data quality, governance and acceptance not only waste resources, but also squander trust - among management and employees. The EY European AI Barometer 2025 shows that 70 percent of German employees fear job losses due to AI, while 36 percent are worried about their own jobs. The key question is therefore: does AI replace or support?

This is why a clear stance is needed: agentic AI is neither an end in itself nor a panacea, but a tool for reducing workloads, stabilizing processes and speeding up decision-making - but only if it is used thoughtfully and strategically in line with clear value propositions. In my opinion, four factors are decisive here:

1. Enabling a technological fit: Agentic AI requires systems that integrate data, map processes in a modular way and can be flexibly expanded - for example via standardized interfaces for AI modules. Whether on-premise or in the cloud - the ability to integrate into existing IT landscapes is crucial.
2. Pick up specialist departments: It is not the executive floor or the IT department, but the specialist departments that know the processes, levers and weak points within an organization best. They need to be involved at an early stage - through clear roles, training, change management and practical pilot projects.
3. Create trust and security: Human responsibility remains, especially for security-relevant or business-critical decisions. Trust in Agentic AI is created through comprehensible decision-making processes - for example through decision logs, source references or visual representations of the chain of reasoning.
4. Ensure knowledge transfer: If a third of skilled workers retire within five years, it is imperative that the knowledge of experienced foremen is transferred to AI-based knowledge systems. Otherwise, companies will lose valuable expertise - and promising agentic AI projects could be on the brink of collapse by 2027.

Those who make targeted investments now - with clearly defined projects and realistic expectation management - can operationalize Agentic AI in an economically sensible way. It's about making smart decisions, not blind trust