The article examines how Large Language Models (LLMs) such as ChatGPT, Claude, Gemini and Copilot can support the software development process - in particular specification, implementation and testing. For this purpose, the development of an astronomical web application is carried out, which supports amateur astronomers in planning their observation evenings by calculating optimal observation times. The practical scenario combines manageable complexity (LLM-compatible) with technical depth: coordinate systems, time conversions and visibility calculations require domain-specific knowledge. The lecturer contributes his own experience as an amateur astronomer and has already developed comparable tools manually.

The focus is on:

• the quality and correctness of the specifications, codes and test cases generated by the LLMs,
• the traceability, robustness and maintainability of the generated artifacts,
• a systematic comparison of the models based on defined QA criteria.

AI - the new miracle component that promises a lot, including in test automation. Many tools claim to make testing more efficient and robust: self-healing, automatic test case and test data generation, intelligent error analysis. Sounds impressive - but does it really work in practice? How easy is it to use the tools? How do they influence the way we work? What opportunities do we see? What risks can the use of AI-supported tools entail? Can we trust them blindly?

We would like to answer these questions by analyzing existing AI-supported solutions.

The presentation will highlight the practical use of Large Language Models (LLMs) in combination with the Model Context Protocol (MCP) for the efficient creation and maintenance of automated test scenarios in development projects. It demonstrates how MCP, as a standardized interface, enables the integration of LLMs with external data sources, tools and test frameworks such as Playwright, thereby reducing manual, repetitive tasks in test scenario development.

GitHub Copilot's Agent Mode enables the autonomous execution of complex development tasks and offers significant efficiency and quality benefits in the regulated medical environment - but requires clear boundaries in terms of control, responsibility and regulatory security.

AI tools like GitHub Copilot are transforming how developers work-boosting productivity by up to 55%. But where do these tools truly shine, and where do they still fall short? In this talk, we present fresh insights from a developer survey and expert interviews, exploring how AI is used across the software development lifecycle. Join us to discover which phases benefit most, what challenges remain, and how AI can be better integrated into everyday workflows.

Results of our study reveal that 93% of developers use AI during implementation, but satisfaction is moderate (3.52/5). AI is also used in requirements (56%),testing (80%), and documentation (78%), with helpfulness ratings up to 4.46/5. Despite adoption, 53% feel restricted by internal policies, and only 10% trust current legal frameworks to keep pace with AI.

Some things can only be understood once you have experienced them yourself. In France, we don't talk about "disabled people" in the context of accessibility, but about "situationally impaired people". Sounds like politically correct language, I thought - until the ophthalmologist in the emergency room proved me wrong with a terse sentence: "You can't read for a week." Boom - off - no WhatsApp, no Google, not even a quick look at the rain radar... Definitely situationally impairing!

I put up with it for two days, then I had to find a solution. I was saved by the ChatGPT voice interface. At last I was able to look up what I was interested in on the Internet again. What was the first thing I looked for? How to have emails or PDFs read aloud on my smartphone, of course. It works - and I am very grateful to all the developers and testers of accessibility!

This memorable week has taught me a lot about the possibilities and limitations of accessibility with and without generative AI - points that I would like to share in my experience report. Because I have learned something else: accessibility is not a goal, but a path worth taking. AI will play an important role in this!