Not every problem needs an agent.

A decision matrix for the architecture choice every team makes (often badly): single-shot LLM, RAG, ReAct agent, Plan-and-Execute, or multi-agent? Six common task types crossed with five approaches gives you 30 cells. Click any cell to see why it's a good or bad fit, what it costs, what fails, and when to reach for something simpler. The recurring lesson: most "we need an agent" instincts are wrong, and the simpler approach usually wins on cost, latency, and reliability.