AI Coding & GenAI Meetup München

Wird ins Deutsche übersetzt…

Join us in our office in Munich on **Thursday, October 29th, 2026** for our **AI Coding & GenAI Meetup** – an interdisciplinary gathering for AI enthusiasts and experts from all backgrounds. In this installment of our meetup series, we focus on how artificial intelligence can be used and **how to do it safely**. We will have **two talks** by our team and by an external guest. Alongside the talks, we will also feature live showcases with plenty of time for networking, drinks, and snacks. For the first talk of the evening, our guest Alexander Muresan will talk about how he builds software now and explain his approach for Agentic Coding: building a system or a so called **"Factory"** that **builds and maintains** the system with a built-in **feedback loop**. After a pizza break with plenty of opportunities for networking, we will have a second talk. In our second talk, Maximilian Bode, Partner at TNG, will talk about **Sovereign Agentic AI**. Agentic systems are becoming the default interface for delivering products and services. But the same autonomy that makes them productive also makes them a strategic liability. Maximilian will walk through the **engineering behind trustworthy enterprise AI**: **hardened inference stacks**, **agent harnesses**, **guarded workflows**, and **network controls** that keep autonomous systems productive without letting them quietly phone home. Let's continue to explore the possibilities and challenges of generative AI together! **Agenda** 18:15 – Doors open 18:45 – Talk 1, "Build the system that builds the software: designing a coding factory for AI agents" by Alexander Muresan 19:30 – Pizza break and networking 20:00 – Talk 2 “Sovereign Agentic AI - Trust, control and the road to dark factories” by Maximilian Bode 20:45 – Q&A, discussions, and networking **Abstracts** **Talk 1: "Build the system that builds the software: designing a coding factory for AI agents"** *Alexander Muresan (Co-Founder, AI Agency Osynk, https://osynk.ai/)* Most teams use coding agents one prompt at a time: someone asks, the agent writes code, a human reviews it. That works for small tasks but doesn't scale, and quality depends on whoever wrote the prompt that day. In this talk I'll show a different approach: instead of building features directly, you build the system that builds them. I'll walk through how we design these "software factories" step by step. First, a shared foundation of rules, design system, codebase knowledge and tools that every agent draws on. Then giving agents access to the running product, so they can check their work against the real thing. Then Workflows for repeated jobs like bug fixes, features and experiments, with a check at every step that sends failed work back with the reason. The safety comes from the structure: tests, design checks and an independent review run on every change, and only work that passes all of them ships. People keep the decisions that need judgement. Finally, I'll show how usage and errors from each run feed back into better rules and checks for the next one. **Talk 2: "Sovereign Agentic AI - Trust, control and the road to dark factories"** *Maximilian Bode (Partner, TNG Technology Consulting GmbH)* Agentic systems that plan, act, and execute on our behalf are becoming the default interface for delivering products and services. But the same autonomy that makes them productive makes them a strategic liability. Risks can be found in poisoned training data, the biases of foreign model and infrastructure providers and Sleeper Agents that lie dormant until triggered. The threat surface of agentic AI extends well beyond classic cybersecurity and into the territory of geopolitical and supply-chain risk. Sovereignty over the models we use, the hardware they run on, and the workflows they touch is the precondition for trustworthy enterprise AI. Openly available model weights and compatible APIs have removed the hard technical blockers of just a few years ago, but production-grade sovereign deployments still takes real engineering. We will talk you through the details behind hardened inference stacks, agent harnesses, guarded workflows, and network controls that keep autonomous systems productive without letting them quietly phone home. Finally, we look ahead to a future of unattended agents and "dark factories" of business delivery - and to the opportunities this shift creates.