I think good AI development is less about simply adding an LLM to an application and more about building a system that actually solves a user problem reliably. The interesting part for me is everything around the model — structuring the data, designing the workflow, using RAG when the model needs reliable context, handling tool calls, evaluating outputs, and making the experience simple for the user.I’ve found that AI systems need to be treated more like software products than isolated prompts. A great demo can generate impressive answers, but a useful product needs predictable behavior, good failure handling, and a clear way to measure whether the AI is actually helping.That is the direction I enjoy working in: combining software engineering with AI to build things that are practical, not just impressive in a demo.