This series is a practical map for developers who want to move past agent demos and into real agent engineering. It is organized into three connected subseries. The Agent Loop starts with the core execution model and shows how tools, stopping logic, traces, evals, and failure handling turn an LLM into a controllable system. Memory Management in Agents then reframes memory as an architecture problem, not a magic feature, covering state, retrieval, summarization, compaction, write policies, personalization, and safety under real latency and cost constraints. Multi-Agent Systems closes the journey by asking the harder question first: when does decomposition actually help? From handoffs and supervisor patterns to coordination, shared state, conflict resolution, evaluation, and cost control, the series builds toward production-ready judgment. The thread throughout is simple: reliable AI agents are engineered, observed, and constrained, not merely prompted.