Which AI architecture should you actually build?

You have a problem and a vendor pitching you their architecture of choice. This guide gives you a five-question decision tree to pick between six common AI architectures: a single LLM call, single agent with tools, multi-agent, long context, RAG, and fine-tuning. Each branch includes a representative use case, cost ballpark, and the three failure patterns we have seen most often. Comes with three rules of thumb for when none of the six fit.