Agent vs Workflow vs Chatbot: Most "Agents" Are Not Agents - Episode 3 of Designing Agentic Systems

Every company right now is running an AI pilot, and the word that comes out of every one of them is "agent". Very often it is not an agent at all. There is only one question that separates a chatbot, a workflow and an agent, and it is not how clever they are. It is who decides the next step. In a chatbot the human decides. In a workflow the developer decided, months ago, at design time. In an agent the model decides, at runtime. Get that one question right and you will choose the correct pattern every time. This is Episode 3 of Designing Agentic Systems, a series on building AI agents the way an architect would. In this video: · Why the agent and LLM loop is slow, and what non-deterministic actually means · The one question that separates all three patterns · Chatbot, workflow and agent, defined in plain language · Mumbai to Pune, answered three ways, so you can feel the difference · A loan journey as a workflow, the way banks already build them · Why a fund transfer should not be an agent, however modern that sounds · So where do agents actually earn their place? · A failed transaction, investigated by an agent, step by step · The thin line: how an agent sits inside one step of a workflow · Chatbot, human decided. Workflow, developer decided. Agent, model decided. NEXT IN THE SERIES Episode 4 — Anatomy of an Agent: Reasoning, Memory, Tools, Orchestration I'm Ehson Golandaz, an enterprise architect with 23+ years in banking and financial services. This channel is about how modern systems, including AI systems, actually get built. We are working towards one thing across this series: a working agentic AI system for an enterprise. Not a demo. Something that survives production, audit, and a regulator. New video every week. #AIAgents #AgenticAI #AIWorkflow #SystemDesign #EnterpriseArchitecture