What Enterprise Architecture Taught Me About Building AI Systems

I keep hearing that AI development requires entirely new skills. New frameworks, new paradigms, new ways of thinking.

Maybe. But the skill I use most when building AI systems isn’t prompt engineering or model tuning. It’s enterprise architecture.

When I’m designing an agentic workflow, I’m doing the same thing I’ve done for years: decomposing complex problems into manageable components. Defining clear interfaces between them. Documenting how information flows through the system. Thinking about what happens when a component fails or needs to be replaced.

These AI systems have more moving parts than traditional software — models, prompts, retrieval strategies, agent definitions, tool integrations, context management. Without architectural discipline, it turns into a mess fast. I’ve seen it happen. Brilliant prototypes that nobody can maintain, extend, or even explain three months later.

The irony is that the people best equipped to build sustainable AI systems might not be the ones who know the most about AI. They might be the ones who know the most about building systems, period.

I’m not saying domain knowledge in AI doesn’t matter. It does. But if you have to choose between someone who understands transformers inside-out and someone who understands how to architect complex systems — for production work, I’d take the architect every time.