Building a Workflow You Will Keep

Most people plateau in the same place: they can get good output when they concentrate, but it has not changed how they work. The gap between those two states is a workflow — a small number of tasks where you always use AI, always the same way.

Start from your week, not from the tools

Write down what you actually did last week and how long each thing took. Then find the tasks that are frequent, mechanical, and low-stakes when wrong. That is where to start — not with whatever is most impressive.

• Frequent — runs several times a week, so a small saving compounds
• Mechanical — follows a pattern you could describe to someone else
• Low-stakes — a wrong result is annoying, not damaging
• Currently unpleasant — the tasks you postpone are the ones where speed changes behaviour, not just time

Turn a good prompt into a template

When a prompt works, do not let it disappear into chat history. Save it with the variable parts marked, so next time you fill in blanks instead of rewriting from memory.

This is the single highest-return habit in the whole course. It converts a one-off good result into a repeatable one.

Write a {{length}} reply to the customer message below.

Audience: {{who they are}}
Outcome I want: {{what should happen next}}
Facts they need: {{price / timeline / policy}}

Rules:
- Plain tone, no marketing language
- One clear next step
- Do not promise anything outside the facts above

Customer message:
{{paste}}

Decide what you delegate and what you keep

This is a judgement about your own value, not about capability. The rule that holds up: automate what your clients or colleagues do not pay you for; keep the part they do.

If people value your writing, use AI for the research, the outline and the editing pass — not for producing the prose. If people value your analysis, use it for gathering and formatting, not for the conclusion.

Measure, honestly

After two weeks, ask whether the task actually got faster including verification and editing — not whether the first draft appeared faster. Those are different numbers, and only one of them matters.

If a task did not get meaningfully faster, drop it and try a different one. That is not failure; it is how you find the tasks that suit this.

Chaining tools into a real workflow

Once individual tasks work, the compounding comes from connecting them: research feeds the outline, the outline feeds the draft, the draft feeds the social posts, the transcript feeds next week's content.

Each handoff is where things break, so keep them explicit. Know what leaves each step and what the next one needs. A workflow that only exists in your head stops working the week you are busy.

Where to go from here

You now have the model, the prompting habits, the grounding technique, the verification discipline and the workflow frame. That is genuinely most of what matters — the rest is practice on your own work.

Two suggestions. Build your own prompt library from what works for you, rather than collecting other people's. And re-run this course's exercises in three months on the model you are using then; the habits transfer even as the tools change underneath them.