Pick one of the five.
Not all five. One. The one whose pain you feel most clearly today. If the inbox is your daily cost, build the inbox router. If Friday afternoon is the one you keep losing to competitor research, build the brief. The third option is the bookkeeping intake, which is the one to build if your accountant calls you on Wednesday with three questions you should already have answers to.
Build it. Set up the folder, write the prompts, run it for two weeks. The first week will be calibration; the prompts will be wrong, the routing rules will miss a contact, the template will ask the wrong question. The second week, the workflow gets useful. By the third week, you stop noticing it, which is the sign it’s working.
Don’t try to automate the company at once. The temptation, after one workflow works, is to build three more on the same weekend. The risk is that you build four half-working workflows instead of one solid one and three planned ones. Operator-grade AI work compounds slowly. One workflow per month, well held, beats five workflows per month half-held.
When you build, hold the four lines.
Structure before automation. Spend the time on the folder, the prompt file, the template, and the rules file. The model isn’t what makes the workflow work; the structure is.
Draft before action. A workflow drafts. A human acts on the draft. The boundary between draft and action is the line a workflow shouldn’t cross by itself, even when it could.
Review before authority. Posting, sending, signing, deciding. These are acts of authority and they belong to people. Workflows compress the work that leads up to them, but they aren’t the ones who carry them out.
Memory before scale. A workflow that runs once is a thing; a workflow that runs for six months is a different thing. The difference is the memory the company built around it: the rules file the operator kept editing, the calibration notes that piled up on the side, the conversations the workflow forced between people who used to do the work alone. Build that memory as you build the workflow, or you’ll lose the workflow when the operator leaves.
That’s the series. Five workflows, one capstone, four lines. The operator’s job is the same as it always was: decide what’s worth doing, set up the conditions for it to happen, watch what actually happens, and adjust. AI doesn’t change the job. AI changes how much of the job’s preparatory work a small team can do, which changes how much of the actual job a small team can take on, which is the only kind of leverage that ever mattered.