Workflow or agent?

At the end of Tool calling, Larkspur’s assistant could look up orders and issue refunds, through MCP servers, behind an approval gate and a limit. Each request was one small job. Now the support lead wants to hand it whole cases: “this printer arrived cracked, sort it out.”

This chapter: Who should pick the steps for a case: code written in advance, or the model?

You will learn

The difference between a workflow, where code picks the path, and an agent, where the model does
Why a workflow is cheaper on a case its authors foresaw
Why an agent copes when a case takes a turn nobody wrote a step for

Words you’ll meet (3)

Goal: What the case should end with, such as “the customer has a working printer and knows it”. It says what, not how.
Workflow: A fixed path written in code: the same calls, in the same order, every time. Larkspur’s team calls it the script.
Agent: A model that works toward a goal by choosing its own next step, again and again, from what each tool result tells it.

The picture on the left is live. Each step below asks you to do one thing in it; the story continues here as you go.

Step 1 of 8

The support lead starts handing the assistant whole cases. Larkspur’s team already wrote a script for the routine ones: the same calls, in the same order, every time. That is a workflow. The other way is an agent. Anthropic calls a system a workflow when code paths written in advance orchestrate the model and its tools, and an agent when the model directs its own process and tool use. Read the first case.