My article on AI-assisted Process Visualization is now available in print and as E-Paper: iX-Magazin issue 03’25. Its pre-release on heise+ has sparked a remarkable amount of feedback from readers worldwide. As questions went beyond the scope of the initial publication, I decided to compile a Q&A in English to share select follow-ups.
This Q&A covers topics such as: 💼 Enterprise-oriented AI frontend 🥟 Chinese LLMs: Qwen & DeepSeek 🙏 Why I still say “please” in my prompts …and more.
In case you missed it: my article has the overarching theme of “AI-assisted Collaboration”. Building on established diagramming methods, my piece shifts the focus to harnessing Generative AI to streamline tasks, detailing along the way on 9 pages: 🦾 Compliance-driven Human-AI collaboration 🔍 Vendor selection and product review 💰 Realistic cost-considerations 👩🏻🏫 Insights from cutting-edge AI/ML research ⚠️ Common pitfalls, including RAG 🗺️ Advanced diagram types from unstructured data
Throughout, I share hands-on methods for an AI-first practice: how can mindful use of AI fuel personal productivity - without compromising trust or burning budgets. I draw from real work experiences and recent research, highlighting what truly works in production – and which pitfalls remain.
In contrast to traditional work styles, this AI-assisted approach provides work artifacts, e.g. diagrams, as something more ephemeral: rapidly generated, fit-for-purpose elements that can be quickly iterated or recreated as needed, using natural language.
Heise+ Article “Hands-on: Process visualization with generative AI” (available in German language: “Prozessvisualisierung mit generativer KI im Praxistest“) (€)
While iX caters primarily to German-speaking IT professionals, the concepts discussed may also be of interest to those with a focus on Business and/or legal or other disciplines. Happy to discuss and clarify.
Particularly the fact-sheets reflect the research status as of late December 2024. To address the challenge of publishing print content in this fast-moving field, I’m maintaining an updates document on GitHub to keep the information current.
[Update 2025-03-13] To those wondering why the somewhat older GPT-4 Turbo still appears - despite its known limitations: essentially, it serves as a placeholder for the now officially released GPT-4.5. The new model excels particularly when creativity and high-detail generation are required. In fact, for the wireframe example discussed in the article, GPT-4.5 even introduced a visualization trick that was new to me: it put the user interface to be presented in a container shape resembling a browser window - which is a predefined, built-in shape: “mxgraph.mockup.containers.browserWindow”. Echoing the spirit of Microsoft’s earlier research paper “Sparks of AGI”: the sparks are indeed flying denser now!
Cost-wise, GPT-4.5 isn’t as steep as some commentators feared; measured precisely, it’s around 5 to 17 times pricier than the AWS option presented in the article. However, as explained in my cost considerations, this difference might well be negligible in practice. Prompting GPT-4.5 does come with its quirks, though: vague, unspecific queries will yield equally simplistic answers. But by using the prompting techniques detailed in the article, activating its deeper knowledge clusters works exactly as intended.