Generative AI and Agentic Architecture in Engineering and Manufacturing

Introduction: The Transformation of Engineering and Manufacturing through AI

The integration of Artificial Intelligence (AI) and Generative AI (GenAI) in engineering and research & development (R&D) offers significant potential. Ködding and Dumitrescu emphasize that AI enables fundamental changes in strategic product planning by utilizing new technologies and data processing approaches to design complex systems [7].

Although GenAI is currently in the spotlight, it is only a part of the broader AI spectrum. Disciplines such as Machine Learning (ML), Natural Language Processing (NLP), and Data Science form the foundation of modern AI applications and play an integral role in industrial processes (Figure 1). However, GenAI expands this spectrum with capabilities for efficiently processing unstructured data, reasoning, and knowledge management beyond traditional data analysis.

Figure 1

The AI Continuum: Positioning Generative AI within AI Disciplines and Supporting Foundations such as Data Management and ML-Ops (Accenture, 2024, modified)

Studies highlight that by 2025, GenAI will be used in 30 percent of companies for AI-supported development and testing strategies [3]. By 2026, over 100 million people could collaborate with AI-driven “AI colleagues”, and 75 percent of companies plan to use GenAI to generate synthetic customer data.

However, the introduction of AI and GenAI in companies presents significant challenges. According to a 2022 study by Accenture, only 12 percent of companies worldwide can utilize AI and integrate it into their business processes effectively [1]. More than 60 percent of the surveyed companies are still experimenting and struggling to scale AI solutions sustainably [1].

These figures illustrate the complexity of implementation and the need for targeted strategies to integrate AI into business processes successfully. The lack of access to large, high-quality datasets, the requirements for explainable, accurate, and reliable AI processes, and the integration into existing, often complex IT landscapes complicate widespread integration. At the same time, there is often a lack of standardized interfaces to seamlessly incorporate AI technologies into existing systems and IT solutions along the value chain.

By training on enormous datasets and flexibly adapting GenAI applications using various techniques, this technology has the potential to make development and manufacturing processes along the entire value chain more efficient, adaptable, and agile.

Despite these hurdles, a clear trend is emerging: the integration of AI and GenAI has the potential to sustainably transform not only engineering and manufacturing but the entire value chain. These technologies enhance productivity, optimize processes, and enable informed, data-driven decision-making in real time – a crucial competitive advantage for companies that embrace this change early on.

The following analysis highlights the potential of these technologies in the engineering and manufacturing landscape, which use cases are already being successfully implemented today, what architectural foundations are necessary to fully leverage their potential, and what companies need to do to sustainably benefit from this technology.

Status Quo: AI in Engineering and Manufacturing

Although AI is increasingly used in manufacturing and engineering, the adoption rate and penetration vary. According to a study commissioned by IBM, 32 percent of German companies with over 1,000 employees actively use AI, while another 44 percent are currently experimenting with AI or exploring its use [14].

In the manufacturing sector, AI is primarily used to optimize production processes. Machine learning and data analysis enable companies to automate production workflows and increase efficiency [5, 10]. Particularly in the automotive industry and electronics manufacturing, there are numerous applications where AI is used to predict machine failures and improve product quality [8]. The use of AI in many small and medium-sized enterprises in the DACH region is limited, as the necessary resources and expertise to effectively implement AI systems are often lacking [10].

In the area of R&D, a similar picture emerges. AI is increasingly being used as a supportive tool in generative design and research to analyze data and identify patterns crucial for developing new products or technologies [12]. Nevertheless, integrating AI into research processes often remains fragmented and heavily dependent on the respective discipline. In engineering, for example, AI is primarily used in design and simulation, while in other areas, such as materials research or product development, its application is not yet widespread [8].

In summary, the use of AI in engineering, R&D, and manufacturing in the DACH region is already established in specific disciplines along the value chain. However, comprehensive and widespread implementation is still pending.

Generative AI in Engineering: Potential and Main Use Cases

Generative AI represents a promising extension of existing methods in engineering and manufacturing. Okuyelu and Adebayo demonstrate that companies can reduce costs through GenAI, shorten time-tomarket, and enhance their innovation capabilities [3, 11].

Large language models (LLMs) are evolving into powerful tools that go far beyond their original application in text processing. They can process classic textbased inputs and multimodal data sources such as images, videos, speech, structured production data, unstructured documents, and technical formats like CAD models (Figure 2). This diversity makes them particularly suitable for complex industrial use cases.

Figure 2

The Diagram shows how GenAI integrates various Data Types to generate usable Results (own representation)

In current industrial applications, the input data comes from various sources, including maintenance logs, ERP systems, technical manuals, inspection images, production data, software code, and 3D engineering data. GenAI systems can analyze these data, link them together, and derive informed insights.

The resulting outcomes include not only text content but also a variety of applications such as optimized production schedules, CAD/CAM designs, formulations, predictions of machine failures or quality deviations, proactive recommendations for action, and process optimizations, for example, through path planning for robotic systems or efficiency improvements in supply chain performance.

Large language models (Foundation Models) do not operate in isolation but can be further adapted with additional data to address certain domains or use cases. This can be achieved through fine-tuning, a targeted model development with domain-specific data, or specialized input instructions (Prompt Engineering). Furthermore, operational actions can be triggered through interfaces to existing IT systems such as ERP or PLM tools, making GenAI particularly effective in practice.

GenAI Along the Value Chain

Accenture estimates that Generative AI can increase the available resources in research and development along the value chain by up to 36 percent (Figure 3).

Figure 3

How Generative AI complements the Value Chain in R&D and Engineering (Accenture, 2024)

In early development phases, GenAI can provide significant support by acting as an intelligent assistant to facilitate idea generation, regulatory analyses, and technical feasibility studies. Its ability to analyze and structure large amounts of data in real time enables more efficient decision-making and accelerates the transition phase from research to conception.

GenAI optimizes iterative processes in the design and prototyping area by automatically simulating and evaluating various design options while considering requirements and regulations. This automation significantly reduces time and personnel effort, shortens development cycles, and accelerates time to market.

Also, in the testing and validation phase, GenAI unleashes its potential by augmenting traditional physical tests with virtual simulations. Weaknesses can be identified early on, and alternative solutions can be explored before physical prototypes are created. Hybrid approaches that combine simulation-based and physical testing ensure reliable results while also reducing costs

In lifecycle management, GenAI enables continuous optimization of products throughout their entire lifespan through data-driven analyses. Systems like Federated Product Lifecycle Management aggregate data from various phases of the product lifecycle to improve maintenance strategies and enhance product performance. This approach is mainly supported by technologies such as knowledge graphs, multi-agent systems, and cloud-based solutions. They make it possible to create robust federated systems while providing a centralized data view.

In summary, GenAI can become integral to modern engineering and manufacturing processes. By strategically deploying it along the value chain, companies can achieve significant efficiency gains and secure their market position through innovative products and processes in the long term.

Challenges of AI Integration in Engineering and Manufacturing

Despite the potential of AI in engineering and manufacturing, companies face numerous structural and technological hurdles that complicate the implementation and use of these technologies. According to a study by Buxmann and Schmidt, 95 percent of respondents believe AI will play a crucial role in their companies. Yet, the actual implementation lags behind these expectations [4].

The challenges associated with the introduction of AI are diverse, as summarized in the Figure 4. Companies are confronted with increasing complexity in data processing and the need to adapt their business models to meet the demands of digitalization [13, 7].

Figure 4

Central Challenges in the Engineering and Manufacturing Industry: Interdisciplinary Collaboration, Data Fragmentation, IP and Data Accessibility, Tool Diversity, and Requirement Complexity (own representation)

Architectural Foundations: Scaling GenAI in Engineering and the Role of Agent Systems

Integrating AI into engineering and manufacturing processes requires a robust and scalable architecture to overcome existing media breaks between tools and disciplines. Modern processes utilize various specialized software tools that often work in isolation. This lack of interoperability leads to data losses, inefficiencies, cumbersome integration, and difficulty for AI to access relevant information.

AI agents can overcome these media breaks by acting as intermediaries between tools and processes. They integrate data sources such as databases, document management systems, or real-time data streams into a unified digital process chain. By leveraging GenAI and an agentic architecture, companies can achieve higher efficiency levels, reduce operating costs, and improve product quality [6].

In particular, in a multi-agent system (MAS), autonomous agents represent specific machines, tools, or processes and can communicate and collaborate. This capability enables the dynamic optimization of production schedules, resource allocations, and process control. Studies show that MAS systems can significantly increase efficiency and flexibility in manufacturing environments [9, 15].

Furthermore, Okuyelu and Adebayo demonstrate that integrating generative AI and agentic architecture in engineering and the manufacturing industry increases productivity, optimizes processes, and enables real-time data-driven decision-making, providing companies with a crucial competitive advantage [11, 3].

Integrating GenAI into engineering and manufacturing requires a modern architecture capable of efficiently processing heterogeneous data sources and transforming them into valuable insights. This architecture is based on a multistage approach, including data collection (Figure 5), transformation, storage, and analysis.

Figure 5

Modern AI Architecture: from Data Collection through Processing and Storage to Orchestration by Agent Networks – a Scalable Structure for Efficient and Action-Oriented Solutions (Accenture, 2024, modified)

The starting point is the integration of relevant data sources such as databases, emails, or technical systems. The data can be converted into a unified format optimized for AI models through methods like embedding and chunking. Subsequently, combining vector databases and knowledge graphs allows for data exploration and mapping of complex dependencies and relationships. At the end of the process, an orchestrated agent network performs tasks such as predictions, optimizations, or problem analyses and delivers actionable results by integrating external data sources and APIs.

Use of GenAI Agents in in Engineering and Manufacturing

To address the challenges in engineering and manufacturing, a combination of robust reasoning systems and specialized GenAI agents has emerged as a best practice. This model combines the strengths of a central orchestrator with the specific capabilities of individual agents tailored to particular tasks and systems (Figure 6).

Figure 6

Architecture of a GenAI-based solution in engineering, consisting of a central orchestrator for task distribution and specialized agents for handling individual process steps, which collaborate iteratively (Accenture, 2024, modified)

The orchestrator takes on planning, evaluation, and user interaction tasks. It manages the entire process, assigns tasks to the agents, and consolidates the results. In contrast, the GenAI agents focus on specific requirements and interfaces, such as understanding and utilizing IT solutions or processing particular data formats. Furthermore, the agents can consider lessons learned, best practices, and industry-specific and company-specific regulations, such as ISO standards. These agents do not require a comprehensive understanding of the overall process but work with a clearly defined, limited project scope. The results are reported directly to the orchestrator or other agents to solve complex tasks in iterative steps.

This modular concept allows for quick and cost-effective processing by utilizing powerful yet specialized tools for individual tasks. At the same time, the orchestrator ensures that all subprocesses are seamlessly linked and that the results are meaningfully integrated into the overall context. This combination of central control and specialized execution offers a robust and scalable solution that meets the specific requirements of modern engineering processes.

Recommendation: Competitive Advantages through a Company-Wide Digital Foundation

The ability to scale AI projects at the enterprise level is crucial for their longterm success. Companies that only use AI in pilot projects must address the scalability challenge to achieve sustainable effects.

However, using AI and GenAI solutions in practice is rarely possible “out of the box”. For companies to benefit from the potential of these technologies, important prerequisites must first be established:

1. Data quality and availability

   Companies must acquire relevant data and prepare it in a structured form.

2. Technological infrastructure

   Powerful computing capacities, scalable cloud solutions, and the ability to integrate into existing systems through APIs and interfaces are central components for the successful use of GenAI. Additionally, flexible architectures are required to meet the demands of growing data volumes and complex models, as well as robust security measures to ensure data integrity and privacy.

3. Process adjustment

   Internal processes and data flows must be optimized and aligned with the requirements of AI.

4. Staff training

   Employees must be prepared to work with AI and GenAI and be continuously trained. The study by Schönberger shows that successful integration of AI systems into existing workflows requires significant training and adjustment to ensure smooth collaboration between humans and machines [12]. This need for training and adjustment is often seen as one of the biggest hurdles companies must overcome to implement AI successfully.

5. Responsible AI

   Companies should ensure that using artificial intelligence complies with ethical, legal, and social standards. This includes developing transparent models that ensure traceability and fairness and protecting sensitive data through robust security measures. Equally important is continuously reviewing AI systems to minimize unintended biases and risks.

6. Clear use cases and business value

   Pilot projects should be purposefully selected and designed to be scalable to other areas of the company. Use cases that address specific challenges and create clear value for the company should be prioritized. Central management of use cases can also create synergies, avoid duplication of effort, and ensure strategic alignment.

Conclusion and Outlook

Generative AI has the potential to transform engineering and manufacturing processes fundamentally. GenAI opens a new level of efficiency, flexibility, and innovation from conception to design to production and lifecycle management. Companies can overcome existing media breaks by integrating multi-agent systems, knowledge graphs, and modern platforms and establishing seamless digital process chains.

However, the successful introduction of generative AI requires careful planning and strategic decisions. Companies must:

1. Structuring data and processes and aligning them with the requirements of AI.

2. Start pragmatically by implementing use cases with measurable added value.

3. Train employees to build the necessary knowledge for dealing with new technologies.

4. Think long-term by establishing a scalable architecture and a robust digital foundation.

New technologies such as multi-agent systems will further expand the possibilities of GenAI. Companies that act now are laying the groundwork for a successful and innovative future.

Summary

Integrating Generative AI and Agentic Architecture offers transformative potential for engineering and manufacturing processes. With its ability to process and analyze complex, multimodal data, Generative AI enables real-time decision-making, predictive maintenance, and process optimization across the value chain. This approach addresses challenges such as fragmented data, intellectual property constraints, and interdisciplinary collaboration. Businesses can unlock significant efficiency gains and enhance competitiveness by leveraging a robust digital core and modern architectures. Generative AI reshapes traditional engineering practices, setting the foundation for scalable, sustainable innovation.