ITCS Hamburg

Agents without oversight do not represent progress – they are a risk. Over 80% of the AI agents in use today operate entirely without a policy framework or governance structure. For IT and security teams, this means uncontrolled processes, unclear lines of responsibility and growing compliance risks, particularly in areas where automation is actually intended to ease the workload.

This presentation demonstrates why autonomy and policy conformity are not contradictory, and how compliance teams can actively ensure AI agents are not hindered but truly ready for deployment. From policy to productive guardrails: a practical look at AI governance as a strategic lever for secure, scalable agent infrastructure.

Takeaways:

• Why uncontrolled agents are an underestimated security risk
• How governance structures increase efficiency rather than reduce it
First steps from policy document to technically implemented guardrail

Large language models only reach their full potential when they are applied to relevant information. Company knowledge be accessible. In this lecture, Melena Thieß will provide a concise overview of what so-called Retrieval-Augmented Generation (RAG) Agents are, how they work, and why they currently play a central role in modern AI solutions. Using examples from the Everyday consulting shows Obungi how they develop smart AI systems as an IT consultancy.

Banks are increasingly relying on AI – but how do you roll out tools like Gemini Enterprise and AI in coding in an environment where every answer can have regulatory consequences? This presentation shows how Berenberg is approaching AI enablement in practice and makes one thing clear: the promises of hyperscalers are not enough – only verification transforms AI from a risk into a reliable tool. Verification is the prerequisite for end-users to truly trust AI in their daily work, making it usable and helpful.

Vibe Coding is considered the great equalizer: no more coding knowledge needed. AI does the rest. Finally, tech for everyone.

But is that true?

Germany is currently funding a nationwide study – tech2STAY, funded by the Federal Ministry of Research, Technology and Space – to understand why women are leaving the IT sector. 2026. We still don't know.

In this keynote, Dajana Eder takes apart – and reassembles – the promise of Vibe Coding. Not as criticism. As an argument. Because Vibe Coding delivers on its promise. But only if we actively decide for whom.

Together with the Free and Hanseatic City of Hamburg, JUST ADD AI has proven for over 2 years that AI can also deliver real added value in the context of sustainability. The result: an agent-based RAG system that supports Lisa Eglhofer and her team in making Hamburg's progress towards the 17 global sustainability goals visible and traceable in a data-driven manner. And all this without manual research.

From Harness Engineering to Agent Orchestration

This talk addresses the question: When and why is a specialised legal AI workflow more valuable than a broad, horizontal AI model? It focuses on the requirements that truly matter in legal processes, such as error tolerance, explainable reasoning, verifiable sources, and seamless integration into existing workflows. The presentation demonstrates why the real added value often lies not in the general model itself, but in how AI is structured, controlled, and made demonstrably usable for a specific legal application.

bettersoon wanted to build a product that could generate a suspected diagnosis from symptoms in sick animals, giving pet parents a reliable recommendation even in the middle of the night, even when all phones are off. So they chose Machine Learning. Then they lacked data. The team looked for prevalence values that didn't exist anywhere. And when they did exist, the quality and quantity were insufficient. Lola Kotte spent a year manually building cases, wondering at night if it all made sense and what would happen if nobody wanted it in the end. The team brainstormed together: Where are the sick animals, really, and what can they take off vets' plates so that they'll give them their patient data? Another pivot: an LLM that gave nice answers but asked the wrong questions.

Today, something is unfolding that looks quite different from the initial plan. Lola Kotte, veterinarian and founder, explains what really happened between "we're building a model" and "it's being used in real practices", and what startups continuously need to do to find the right mix of "Compound AI".

The digital transformation of the world of work is exacerbating social inequality, as nearly 40% of EU citizens still lack access to the necessary digital skills. The ReDI School of Digital Integration is tackling this challenge head-on in Hamburg with a new learning platform that provides accessible training in AI awareness and practical AI tools for job seekers. Its offering also includes free specialist IT courses in machine learning, Python, data analytics, cloud computing and cybersecurity, specifically aimed at people from migrant backgrounds and those facing barriers to education. Find out how they are building a bridge to the local labour market through practical workshops and networking with Hamburg-based tech companies. You are invited to become part of this mission – whether as a learner on the path to an IT career or as a volunteer mentor and partner, working together to shape an inclusive tech ecosystem for Hamburg and the whole of Germany.

Roles are shifting. Product owners are writing code. Founders are shipping features. And developers who embrace AI are building things that would have taken entire teams before. In this live demo, Marvin Vortkamp will open the Itemary codebase, make a real front-end change using OpenAI Codex inside VS Code, and show the result running on screen. No prior coding experience required on his end. If he can do this, the bar for what developers can achieve today has never been higher. This session is not about replacing engineers. It’s about what becomes possible when everyone moves one step closer to the code.