Stop Building AI Agents for Everything

Everyone is building AI agents. But a lot of AI products don't need one. A refund request, a support ticket, a document generator: most of them follow steps you already know. This video explains the real difference between a workflow and an agent, which comes down to who controls the next step: your code or the model. We walk through the five workflow patterns from Anthropic's Building Effective Agents: prompt chaining, routing, parallelization, orchestrator-workers and evaluator-optimizer. Each one comes with a concrete example. Then we look at when a real agent is worth its cost, latency and risk, how to combine agents with workflows, and three rules for running an agent in production. CHAPTERS 0:00 Do you actually need an AI agent? 1:01 What is a workflow? 1:35 What is an agent? 2:45 Why not use agents for everything? 3:56 The augmented LLM 4:34 Prompt chaining 5:13 Routing 6:01 Parallelization 6:46 Orchestrator-workers 7:36 Evaluator-optimizer 8:13 When you actually need an agent 9:05 Combining agents and workflows 9:47 Making agents production-ready 10:46 Who should control the next step?