"AI engineer" and "AI architect" get thrown around almost interchangeably in job postings, which is a shame, because the distinction is actually useful once you sit with it for a second.
AI engineering is implementation. Fine-tuning a model, writing the RAG retrieval logic, wiring up an API, making one specific component work correctly and efficiently. It's close to the code.
AI architecture is decisions about shape. How many models does the system actually need, do agents talk to each other directly or through a broker, where does state live, what happens the moment one component falls over. It's closer to the whiteboard than the terminal, even though, in practice, most people making these decisions also end up writing the code that implements them. Nobody hands you a diagram and walks away.
Take Mycellium, the agent-swarm system I led through the TUM × SAP practical course. The engineering question was "how do I get agent A to call the right tool." Fine, solvable, a Tuesday afternoon problem. The architecture question was the one that actually shaped the project: should there even be a central orchestrator, or does a swarm without one produce better behavior here? We went with no central controller. Instead, agents self-select into tasks through capability- and trust-weighted bidding, and consensus gets gated behind a citation threshold via a blackboard protocol, rather than one "boss agent" making the final call. That decision came from the same instinct I mentioned in my journey post: the one-boss-agent pattern felt like the wrong shape the moment agents needed to specialize and trust each other's outputs instead of just reporting up a chain.
Most people doing real AI work do both of these things, engineering and architecture, in the same afternoon, sometimes the same hour. The formal titles, "AI Engineer" and "AI Architect," tend to describe seniority and scope of ownership more than they describe which mental mode you're actually in at 2pm on a Tuesday. A junior engineer can make real architecture calls on a side project. A senior architect still writes code, whether they admit it in interviews or not.
I don't have "AI Architect" on a business card, and I'm not going to pretend a master's student should. But that's the direction I'm building toward, and the split above is the mental model I already use on every project I touch, Mycellium included. Title or not, it's the muscle I'm training.