Software architecture in the age of AI: Why friction is a feature

18. November 2026 Vincent Software-Architektur KI-Ethik Digitale-Souveränität xynap

# Software architecture and human responsibility

Modern software development is undergoing profound change. The massive use of artificial intelligence, especially in the generation of code, pushes the boundaries between automation and manual control. In a live discussion on software-architektur.tv, this paradigm shift was illuminated: friction is not a failure state, but an essential feature of a robust architecture.

Traditionally, software architecture is often understood as the pursuit of smooth execution. However, in complex systems that are increasingly influenced by AI models, a too smooth surface often leads to dangerous naiveté. When a system no longer feels resistance – whether through data inconsistencies, unforeseen inputs or ethical dilemmas – it often lacks the mechanism for self-reflection and correction. Friction here serves as a sensor for anomalies.

This becomes particularly relevant in the context of current debates about digital sovereignty. For example, Irish data protectors recently forced AI provider X to stop training Grok with third-party data. This shows that technical systems cannot operate in a vacuum. An architecture without friction would possibly ignore such regulatory intervention or ethical conflicts until it is too late. The ability of a system to respond to external signals and internal contradictions is a matter of resilience.

In the area of public administration, too, the danger of over-automated, smooth processing becomes clear. The Federal Government plans to use AI in migration procedures. Critics and organizations, however, warn that a pure efficiency orientation hides human nuances. When an algorithm marks and corrects each deviation as an error, without taking into account the underlying human responsibility, a dangerous rigidity arises. Friction allows people to intervene in the process when automation reaches its limits.

The dependence on external technology is another example. European players are actively seeking alternatives to US military software such as Palantir. This search is not only a question of geopolitical orientation, but also of an architectural decision for sovereignty. Systems based on proprietary opaque stacks often do not offer a way to adapt them to local conditions. An architecture with friction detects such dependencies and asks the developer or administrator to consider alternatives before a critical error arises.

The use of AI-generated code also brings new challenges. While AI models can produce efficient code, they do not always understand the context or ethical implications of a decision. A system that interprets and displays these uncertainties as friction rather than ignoring them is safer. It forces the architect to make decisions and take responsibility.

In summary, in a world characterized by AI and networked systems, friction is the protection mechanism against automation errors. It ensures that human intelligence remains in control and that systems do not blindly respond to data without understanding it.

## What this means for users

Xynap’s integrated AI agents and self-hosted infrastructure provide a platform that ensures human control over AI processes and supports team architecture decisions.

Sources (2)

  1. www.heise.de
  2. netzpolitik.org