AI in software testing – hope, hype and liability: an expert’s view from the field

Artificial intelligence is currently rapidly changing how software is developed and tested. Test generation, automatic evaluation, self-healing test systems, and, and, and. Much of this sounds like a liberating blow for quality assurance and engineering.
But what happens if these systems fail?

What does “quality” mean when decisions are no longer made deterministically but probabilistically?
And how does responsibility change when errors can no longer be clearly attributed to people or code?
This presentation combines the perspective of a software expert with the question of the future of testing in the age of AI.

The presentation is aimed at all those who understand testing not only as a process, but also as a security and trust factor and who want to know how quality remains resilient in an AI-driven world.

Implementing accessibility strategically and effectively

Digital accessibility is no longer a voluntary additional task, but a legal and organizational obligation for companies and public bodies. In practice, however, accessibility is often only tackled selectively – without an overarching approach that has a lasting effect. The presentation shows how a holistic accessibility strategy can be developed and implemented: from stocktaking (websites, apps, applications) to a clearly defined target state and prioritized action planning.
One focus is on not treating accessibility as a one-off project, but integrating it into design, development, editing and procurement processes. In addition, qualification measures for relevant roles are presented in order to anchor knowledge and responsibility in the organization. Finally, it will be explained how monitoring, reporting and regular further development (e.g. re-audits, key figures, roadmaps) make progress measurable and ensure long-term quality. The presentation highlights both the requirements of public bodies (e.g. BITV, verification obligations) and the perspective of private companies (e.g. BFSG, risk minimization) and is aimed at decision-makers, project managers and specialists who want to establish accessibility in a permanent and structured manner.

AI-based medical devices. A practical example with its requirements for the software development process and conformity assessment

The development of AI-based medical devices presents manufacturers with complex regulatory and technical challenges. The combination of the EU Medical Device Regulation (MDR), the EU AI Act and the numerous standards – especially the version of the new IEC 62304 ED2, which has been released for comment – defines new requirements for the software development process for AI-supported systems. Various special features and processes must be integrated into the software development lifecycle (SDLC) and planned and documented in a targeted manner. The practical example describes the development and regulatory classification of an AI-based alarm system for use in intensive care units. The aim of the system is to detect critical patient conditions, such as sepsis, at an early stage using continuously recorded physiological measurements and to warn medical staff in good time.

This presentation will use the above-mentioned example to highlight how an AI-based medical device is created, which processes are interlinked and which steps are taken for conformity assessment. The draft of the new IEC 62304 ED2 and other regulatory requirements are also intended to show how manufacturers can adapt the software development process accordingly in order to successfully develop an AI-based medical device. Among other things, the integration of the AI Development Lifecycle (AIDL) into the existing SDLC will be clearly analyzed and the individual phases of planning, design and data management for AI-based medical devices will be discussed. The presentation is aimed at software developers, regulatory affairs experts, product managers and decision-makers in medical technology who are involved in the development, approval and quality assurance of AI-based medical devices.

The Era of Private AI: How Local Agents Automate Operations Without Data Exposure [EN]

For most enterprises, the barrier to AI adoption isn’t technology – it’s trust. This presentation showcases how local AI agents are transforming “Business as Usual” by staying behind the corporate firewall. Drawing from Kalmia’s implementation experience, we will cover practical use cases: from an AI Assistant in manufacturing that handles process deviations to automated email agents that draft professional responses based on internal ERP data. We will demonstrate how these agents function as “digital coworkers” that learn from your private data without compromising security. The session provides a roadmap for companies to achieve “quick wins” in automation while ensuring 100%data sovereignty.

Automated benchmarks for an AI agent

At last year’s Net Week, she spoke in general terms about metrics that can be used to measure the quality of the results of generative AI in general and AI agents in particular. This year, she would like to explore the topic in more technical depth. Which metrics can be recorded automatically and which cannot? What does automated benchmarking actually look like? What statistical considerations are required to take account of the non-deterministic nature of generative AI? Answers to these questions with examples can be seen in this presentation. PS: You don’t need to have heard last year’s talk to understand this one.

Strengthening cooperation through MB testing at Lufthansa CityLine

The presentation will show you how Lufthansa CityLine was able to strengthen the cooperation between requirements engineers, developers and testers through model-based testing (MBT). This allows you to focus on the actions that lead to a real improvement in quality. MBT ensures greater transparency and traceability in the test process. Thanks to the tool-supported test case generation, they are able to create the most relevant test cases for the product. Lufthansa CityLine’s practical insights will help you to make your test processes more efficient and sustainable.

Testing Trends: Shift Left, Shift Right or AI – Which path leads to optimum quality?

The testing landscape is changing rapidly: Shift Left promises early error detection, Shift Right delivers production-related insights and AI is penetrating all areas of testing. But which approach is right for your company? Instead of blindly following trends, modern organizations need hybrid testing strategies that fit their individual corporate culture, technology and specific business goals. This session will show you how to develop a customized quality strategy from the variety of current testing trends – for faster releases without compromising on quality. Learn which combinations of classic and innovative approaches work in practice and how to find the optimal testing mix for your specific challenges.

Evaluation and selection of AI models that can be integrated using APIs for your own solution scenarios

Quality engineering is the new testing

Accessibility: Because the Internet should not be an obstacle course

Quo Vadis quality assurance: between man, machine and magic

Vibe coding for skeptics: Between intuition, misconceptions and real productivity