1. Background and Motivation

The development of complex products involves numerous interdependent activities: stakeholder needs are translated into requirements, system architectures are developed, decisions and assumptions are documented, engineering results are reviewed, and verification and test cases are derived. Although these activities are logically connected, they are often performed by different roles using separate tools and data sources. This results in media discontinuities, incomplete traceability, repeated manual work, and considerable coordination effort.

Generative AI is currently used primarily to support individual tasks. AI agents go one step further: they can pursue goals, plan and execute tasks, access tools and engineering artifacts, and transfer results between different process stages. This creates the prospect of an agent-supported development process covering several stages—from analyzing a requirement and proposing architectural alternatives to reviewing engineering artifacts and deriving test cases.

However, the industrial use of agentic AI requires more than a technically functioning prototype. Companies need reliable reference processes, appropriate agent architectures, clearly defined human decision points, and empirical evidence that the approach improves development time, quality, consistency, or cost.

2. Objective

The objective of this thesis is to develop and evaluate a reference approach for the end-to-end use of AI agents in product development.

The central research question is:

How should development processes and AI agent architectures be designed so that AI agents can effectively and reliably support connected engineering activities from requirements engineering through system architecture and reviews to test derivation?

The thesis should include the following activities:

• Review the state of research on agentic AI, Generative AI in engineering, AI-supported requirements engineering, and AI-supported systems engineering.
• Select and model a representative development process, for example Requirements → System Architecture → Review → Test.
• Identify appropriate agent roles, responsibilities, interfaces, and human-in-the-loop decision points.
• Develop an agent architecture covering orchestration, memory and knowledge access, tool integration, quality controls, and escalation mechanisms.
• Implement a prototype for a clearly defined industrial use case.
• Evaluate the prototype using suitable performance indicators, such as processing time, output quality, defect rate, traceability, rework effort, reliability, and user acceptance.
• Derive design principles, prerequisites, limitations, and recommendations for industrial deployment.

The expected results are:

1. a 3DSE reference process for agent-supported product development,
2. a documented agent and integration architecture,
3. an implemented and evaluated pilot,
4. a KPI framework for measuring technical and business benefits, and
5. recommendations for transferring the approach to additional development processes.

A Design Science Research approach is recommended, with the reference process, agent architecture, and prototype treated as artifacts that are developed, demonstrated, and systematically evaluated. To ensure sufficient depth, the prototype may focus on selected transitions within the overall end-to-end process.

3. Profile

We are looking for a student in Systems Engineering, Mechanical Engineering, Industrial Engineering, Computer Science, Software Engineering, or a comparable degree program.

The ideal candidate has:

• an interest in product development, systems engineering, and Generative AI,
• basic knowledge of requirements engineering, system architecture, verification, or testing,
• an interest in prototyping and evaluating an AI agent solution,
• knowledge of Python, APIs, LLM applications, or Retrieval-Augmented Generation as an advantage,
• a structured and independent working style, and
• an interest in collaborating with industry experts and evaluating technical solutions in a practical setting.

4. Contact

Prof. Dr. Clemens van Dinther

3DSE Management Consultants GmbH

c.vandinther∂3dse.de

5. Starting Literature

• ISO/IEC/IEEE (2023): ISO/IEC/IEEE 15288:2023—Systems and Software Engineering: System Life Cycle Processes.
• NASA (2016): NASA Systems Engineering Handbook, Revision 2.
• Kretzschmar, M. et al. (2024): Evaluating the Current Role of Generative AI in Engineering Development and Design—A Systematic Review. NordDesign 2024.
• Norheim, J. J. et al. (2024): Challenges in Applying Large Language Models to Requirements Engineering Tasks.Design Science, 10, e16.
• Hou, X. et al. (2024): Large Language Models for Software Engineering: A Systematic Literature Review. ACM Transactions on Software Engineering and Methodology.
• Li, X. (2024): A Review of Prominent Paradigms for LLM-Based Agents: Tool Use, Planning, and Feedback Learning.
• Liu, X. et al. (2024): AgentBench: Evaluating LLMs as Agents. ICLR 2024.
• Peffers, K. et al. (2007): A Design Science Research Methodology for Information Systems Research. Journal of Management Information Systems, 24(3), 45–77.