AI Is Just Another System

Look at the products shipping today and you quickly get the feeling that “and throw in some AI!” was ordered right before release. Maybe you’ve seen this in your own projects. But does it really have to be that way, and what does it actually mean for our architecture?

My thesis: from an architecture standpoint, AI is first of all just another system in your landscape. An external, expensive, and non-deterministic one. And increasingly one that acts on its own. In the simplest case, an LLM call is nothing more than an interface. Agentic systems go further: they use interfaces, call other systems, change data. From that perspective, I’ll focus on three aspects in the talk:

– Security: Even without AI, security was often neglected. Now we add attack surfaces that are hard to control, plus external systems that act on their own and access our data. Some of it can be caught with established measures, but far from all of it. – Architecture: What actually changes when this system is non-deterministic and partly acts on its own — for testing, for error handling and fallbacks, for latency and cost, and for accountability? – Necessity: A question asked far too rarely: do we even need the system? Often a simple, classic solution beats the AI add-on, because unnecessary complexity costs more than the feature is worth. Grounded, hype-free, with the goal that afterwards you can decide when AI helps and when it’s just complexity you’re better off without.

Datum

17.11.2026

Uhrzeit

11:15 - 12:00

Konferenz / Veranstaltung

iSAQB Software Architecture Gathering 2026

Ort

Titanic Chaussee Berlin, Berlin