When requirements remain vague, testing becomes a lottery. Across all industries: pure text is rarely enough. Domains differ, but the principles remain the same. Models, use cases, activity diagrams and even PlantUML create structure and a viable test basis. AI provides support as long as quality is systematically ensured. This means clarifying the context, sharpening the scope, defining roles, specifying terms and iteratively obtaining feedback. Confidentiality remains a must. Those who understand requirements as a common model rather than a text document gain speed, security and sustainably better software.
In this episode, I talk to Andreas Günther and Breno Pinheiro about AI for better requirements and a solid test basis. We categorize: Industries tick differently, but without clear requirements, testing becomes a lottery. Text is not enough. Models, use cases, activity diagrams and even PlantUML provide structure. AI helps when we enforce quality: Clarify context, sharpen scope, define roles, feedback iteratively, make terms crystal clear and observe confidentiality. Long, reusable prompts instead of gut feeling.
"The more good requirements there are as a basis for training, the more the AI actually knows what the company is doing." - Andreas Günther, Breno Pinheiro
Andreas Günther is a consultant, coach and trainer at SOPHIST and specializes in language and model-based methods in requirements engineering and systems engineering. He supports customer projects, advises, coaches and trains employees in companies in a wide range of specialist areas. His field of activity includes methods of linguistic analysis as well as requirements models and their combination (STABLE methodology).
As a requirements engineer and trainer at SOPHIST, Breno Pinheiro supports customers in the structured elicitation, analysis, documentation and management of requirements - both in classic requirements engineering and in the context of systems engineering. His particular focus is on the possible applications of artificial intelligence in these disciplines. In training courses and coaching sessions, he teaches how AI can be used profitably, responsibly and in a targeted manner.
Requirements are the foundation of any successful software development. But what makes a good requirement and how can artificial intelligence help teams do a better job? In the podcast with Richie, Andreas Günther and Breno Pinero from the Sophists, we find out how AI applications are already influencing requirements engineering today and what opportunities and limitations they offer.
Many people think of requirements as pages of Word documents or user stories in the backlog. In their consulting work, Andreas Günther and Breno Pinero experience how differently companies work - and how crucial the right approach to requirements is. Whether automotive, banking or healthcare: for a long time, industries had different standards. Today, almost every industry attaches more importance to clean requirements, often triggered by legal requirements and increasing complexity.
Requirements engineering, i.e. the work involved in determining, documenting and checking requirements, is more than just writing specifications. As the experts explain, they spend most of their time gathering the necessary information in the first place. Requirements arise from discussions with stakeholders, workshops and various elicitation techniques such as interviews or brainstorming sessions.
The community initially experimented with the boom in AI tools. However, it quickly became clear that there was a risk of wasting time and resources if questions were asked aimlessly. Value - not quantity - is what counts. The sophists therefore ran through numerous scenarios in which AI can be useful in requirements engineering: from collecting ideas and requirements to structuring a backlog and checking and translating complex requirements into clear, testable specifications.
One particularly exciting area is the generation of requirements texts, user stories or diagrams using AI. Many tools claim that they can create user stories based on any template, for example. However, experience shows that the results are usually only useful if the prompts are formulated clearly and comprehensively.
A good prompt is the key. AI provides many answers to every question - but quantity does not equal quality, says Andreas Günther. To create useful requirements, it helps to describe the context precisely. What is the actual goal? Is it an app, an entire system or just a part of it? Is it a functional or technical requirement? Details such as functionalities, quality aspects (such as performance, security) and the type of documentation also need to be included.
Collaboration with the AI works best as a dialog. Do not expect the end result immediately, but proceed step by step. Let the AI ask whether the context is understood and then refine it together. Supplying a glossary or examples also increases the quality of the AI output. It is important not to allow hallucinations or false sources to be sold as facts - always question critically!
Visual requirements are also becoming increasingly important. Complex relationships can often be better illustrated using diagrams (such as process or activity diagrams). If you ask cleverly, AI can even generate PlantUML code, which is then converted into clear images.
Whether ChatGPT, Gemini or other AI services - according to the guests, there is no clear favorite. All systems are developing rapidly. Nevertheless, caution and the use of secure, possibly internal systems is recommended for confidentiality.
AI can enrich requirements engineering - if it is controlled in a targeted manner. Those who prepare properly, clearly describe the use case and context and remain in dialog with the AI will raise the quality of their requirements to a new level. The sophists also provide guidelines for good prompts and will soon be publishing an entire book with practical instructions and examples. If you start now, you have the chance to lay the foundations for successful projects tomorrow with clearer and better requirements.