A workflow is the ordered path work follows from input to useful output: work flows through a series of granular processes. A process is a sequence of actions, and a core activity is the smallest named piece of work that still matters.
Picture a client file moving across desks in a small office. Someone receives it, someone checks it, someone fills in missing information, someone reviews it, someone sends the finished thing to the client. The file stays the same, but the work changes at each desk.
Why you care
Workflows are training material for AI. Say “handle onboarding” and the agent has to invent the desks. Say “receive the signed agreement, extract these fields, create the client folder, draft the welcome email, then stop for review,” and it has a path to follow; the best skills are really saved workflows. A fixed, rules-based path may only need automation, while one that needs language, judgment, and review fits an agent. Write the path down, run it once, watch where it breaks (missing input, unclear owner, slow handoff), correct it, and keep tightening until the output is boringly reliable.