The reason to name a process is that it turns invisible work into something teachable. It’s the step-by-step layer you can watch, write down, improve, and eventually hand to an agent.
Think about making coffee in a small office. Fill the water. Add the filter. Scoop the grounds. Press the button. Pour. Clean the pot. The outcome is coffee people can drink; the process is the ordered set of actions that gets you there. A process isn’t a system or a workflow. A system is the broader approach to an outcome, a process is the steps inside it, and a workflow is how work moves through several processes, people, and tools.
How it shows up
In client work, a process might be “collect onboarding documents” or “review a vendor bill.” Each has an outcome, steps, inputs, and outputs. For AI, it has to be more explicit than it is for people, because a person skips steps mentally from experience while an agent needs the hidden parts made visible. “Review this monthly report” isn’t enough. What should it check first? Which source files matter? What counts as an error? When should it ask a human? Those details are the process. This is where core activity matters: we break work down until an action is small enough to name clearly, like “compare source documents,” which is far easier to teach than “do accounting work.” A phase is a larger chunk a process sits inside, like an onboarding phase holding document collection, review, and kickoff prep.
Why you care
If you can’t describe the process, you can’t reliably improve it or delegate it, and you’ll keep depending on the person who “just knows how it works.” A process is where work becomes teachable, and teachable work is where AI starts to become useful.