Skip to content
Back to courses

AI at work: from draft to a repeatable workflow

Take one work task from choosing what AI can draft through actions, a repeatable instruction, and a message. Identify where a person makes the decision.

BeginnerAbout 105 min7 lessons
Start course: Choose a task for AI to prepare
  • Choose a part of a work task that AI can prepare, and explain why it is suitable.
  • Turn notes into actions without inventing owners or deadlines, and keep open questions visible.
  • Turn a successful workflow into an instruction with an output and a check at every step.
  • Identify where a person approves the result and what information must not be shared with AI.

Who it's for

For people who already use AI occasionally at work and want to make it a controlled, checkable part of their workflow.

Before you start

Course content

  1. 01

    Choose work AI can prepare

    Start with one work situation and separate out the parts that suit an AI draft.

    1. Choose work AI can prepare

      Break a work task into parts and choose one where an AI draft can help. Keep decisions and confidential information with you.

    1. Choose work AI can prepare

      Do the lesson exercise, then break one of your own work tasks into parts. Mark which part AI will prepare and which parts you will do yourself.

      Check when you're done Each part of your task has an owner and a reason, and people retain the decisions.

  2. 02

    Work with your own material

    Turn notes into a status update and action list while keeping missing information visible.

    1. Status update without invented facts

      Turn notes into a team update where every name, time, and decision can be found in the notes.

    2. Turn notes into actions

      Use the notes from the same situation in the action-list prompt. Separate decisions, owners, deadlines, and open questions.

    1. Status update without invented facts

      Do the guide exercise using the office-move notes, then compare every name and time in the draft with the notes.

      Check when you're done Your update contains no details missing from the notes, and open items remain marked as open.

    2. Turn notes into actions

      Fill in the meeting goal and notes in the prompt. Leave missing owners and deadlines visible.

      Check when you're done The action list contains no invented owners or deadlines, and open questions are marked.

  3. 03

    Make the workflow repeatable

    Turn an action list into an instruction someone else can follow, then shape it into a message for a stakeholder.

    1. A repeatable work instruction

      Build an instruction from the action list, with an action, expected result, and check at each step.

    2. Message for a stakeholder

      Turn the next step in the instruction into a message to a customer or colleague without making new promises.

    1. A repeatable work instruction

      Paste your action list into the prompt and say who the instruction is for and when it will be used.

      Check when you're done Each step has an input, an expected result, and a point someone can check.

    2. Message for a stakeholder

      Fill in the situation, recipient, and key points in the prompt. Keep open items marked as open.

      Check when you're done The message's facts match the action list, and it makes no promises absent from the instruction.

  4. 04

    Keep the decision with a person

    Mark the approval point and decide what information not to give AI.

    1. Where the decision stays with a person

      Take the same situation to the decision-support guide and distinguish a suggestion, a recommendation, and a decision.

    2. What information you give AI

      Remove personal and confidential information from your work material while keeping the task possible.

    1. Where the decision stays with a person

      Do the guide exercise using your own work situation and mark who approves the result.

      Check when you're done You can name the approver, the effects of the decision, and information the AI did not see.

    2. What information you give AI

      Do the lesson exercise, then remove details from your own material that must not be given to the AI service.

      Check when you're done The redacted material is enough to complete the task and contains no identifiable personal or customer information.

Capstone: a repeatable workflow for one work task

Choose one real or fictional work task and remove identifying details. Write a bounded request for AI and check its output against criteria you wrote in advance. Record the result for every criterion. If any criterion is not met, make a targeted correction and check it again; if all criteria are met, record that no correction was needed. Turn the workflow into an instruction someone else can repeat, and mark where a person approves the result.

What to demonstrate

  • A description of the work task and why AI will prepare the selected part.
  • The request given to AI, plus the first and checked outputs.
  • A checklist written before checking and its results.
  • A repeatable work instruction with steps, outputs, and check points.
  • A note showing the approval point and what information was withheld from AI.
  • The check results and, where needed, a targeted correction with a follow-up check; if all criteria were met, record that no correction was needed.

Check your work

  1. Does your description show which part AI will prepare, what you will do yourself, and why?
  2. Did you replace personal and confidential information before giving the material to AI?
  3. Does the request provide enough context and specify the output format?
  4. Did you write the criteria before checking and record each result? If any criterion was not met, did you correct that part and check it again?
  5. Can someone else follow your work instruction, with an input, result, and check point at every step and a named owner?
  6. Does the instruction identify a person to approve the result, and do you know who the decision affects?

Course skills

Teaches

Practiced in assignments