This is the single most important distinction for this audience, and almost everyone gets it slightly wrong, so we make it crisp. The difference between a workflow and an agent is who decides the steps.

A workflow follows steps you defined in advance. You laid out the path, when this form comes in, do these three things in this order, and it does exactly that, every time, predictably. The automation you built in the last lesson is a workflow: you decided the sequence, and it executes it. An agent is different: you give it a goal and a set of tools, and it decides for itself what steps to take to reach the goal, adapting as it goes. You did not script the path; you described the destination and let it find the way. Anthropic frames this exact distinction cleanly: workflows orchestrate steps through predefined paths, while agents dynamically direct their own process and tool use.

Why does this matter to you as a builder rather than as trivia? Because people reach for an agent when a workflow would be simpler, cheaper, and more reliable, and that is a costly mistake. An agent's freedom to decide is powerful when you genuinely cannot predict the path in advance, but that same freedom means more cost, more latency, and more ways for it to go wrong. The professional instinct, and the one this course wants you to build, is to use the simplest thing that does the job: a plain workflow when the steps are known, an agent only when the task truly needs to decide for itself. Knowing which you are looking at, and which you actually need, is a judgment that will save you money and grief.

Hold this line clearly, because the next lesson has you operate an actual agent, and you will want to be able to see exactly where the "it decides" begins.