Most AI projects don’t fail in production because of bad code. You guys can code. They fail because the wrong decisions were made long before the first line was written.

In this talk, Wiebke Apitzsch—Managing Director of AI.IMPACT with a background in strategy, AI implementation, and ethics—takes a DevOps lens on AI transformation. She argues that many teams are asked to productionize use cases that should never have been built in the first place.

Instead of focusing on tools, models, or frameworks, she introduces a practical decision-making approach that integrates four dimensions developers deal with every day—often implicitly:

Tech reality – What actually works beyond the demo? Economic viability – What justifies the infra, latency, and maintenance cost? Adoption & change – Will anyone trust, use, or integrate this into workflows? Ethical impact – What are we automating—and at whose expense?

Using real-world patterns from AI projects, she shows how “category errors with budget” emerge—and why DevOps teams are often the ones left operationalizing flawed assumptions.

The session provides a clear mental model to push back, ask better questions upstream, and turn AI from a fragile experiment into something that can actually survive production.

Because in the end, scalable AI is not a tooling problem. It’s a decision problem.

Wiebke Apitzsch is Managing Director of the consulting firm AI.IMPACT and an expert in the strategic development and implementation of AI solutions in organizations. With many years of experience in