Bring your step plan, context card, and decision record together. A reusable workflow makes the next run easier to review because the stable instructions, changing inputs, and human checkpoints are written down.
“Organize our book swap.”
This leaves the goal, sequence, changing details, and review points unclear.
“Help me prepare a monthly book swap using my saved step plan. For this run, use only these confirmed inputs: [date], [venue], [permission status], [expected attendance]. Follow the context card and chosen response approach. Leave any unconfirmed item blank and ask me what to check. Draft a short invitation, then list its facts for my review. Do not send or schedule anything.”
Stable instructions stay in the playbook; the date, venue, permission, and estimates change each run. A person checks the result before using it.
One complete manual test run
For this fictional book swap, the organizer has chosen optional RSVP. The date and venue are unknown, so they leave those inputs blank. They ask AI to draft a short invitation using the context card, and list every event detail the draft contains.
“Neighbors are welcome at our free book swap. Join us next month! Date: [confirm]. Location: [confirm]. You may bring a book or simply come browse.”
The organizer checks each claim against the inputs. “Free,” “book swap,” and optional attendance fit the example. The date and venue remain placeholders, but “Join us next month” still sounds more definite than the inputs allow.
Human review catches a weakness
Keep a checkpoint before the invitation
The organizer improves the playbook: “If the date is not confirmed, say ‘Date to be announced’ and do not write a date-based invitation.” On a second manual test, the draft uses that wording and lists date, venue, permission, and access details for review. The organizer still verifies those details and decides whether to send the invitation themselves.
This is a manual routine: the person supplies inputs, follows the steps, asks for help at the relevant point, checks the answer, and chooses what to do. It does not run in the background or schedule, publish, or send anything.
One routine · four lessons
Workflow Playbook
Bring your step plan, context card, and decision record together. This playbook is a reusable set of instructions for a person and an AI assistant. It does not run by itself, send messages, schedule events, or make decisions.
Your work saves in this browser and is included in the learning-work export. It does not sync across devices; clearing browser data may remove it. You can use the same headings on paper and carry your notes forward.
Using another AI assistant? Read the beginner’s guide.
Review the work you saved in earlier lessons
Lesson 1 · Workflow Step Plan
- Goal
- Not added yet
- Steps
- Not added yet
- Order And Dependencies
- Not added yet
- Human Review
- Not added yet
Lesson 2 · Reusable Context Card
- Purpose
- Not added yet
- Audience
- Not added yet
- Constraints
- Not added yet
- Desired Format
- Not added yet
- Unknown Details
- Not added yet
Lesson 3 · Decision Record
- Options
- Not added yet
- Hard Requirements
- Not added yet
- Preferences
- Not added yet
- Known Facts
- Not added yet
- Assumptions
- Not added yet
- Missing Information
- Not added yet
- Decision And Why
- Not added yet
Example: event date, confirmed venue, attendance estimate, and supply needs. Use blanks until confirmed.
Write the reusable request in plain language. Tell AI the goal, audience, limits, output format, and what to ask or leave blank.
Bring forward the steps and dependencies from Lesson 1. Mark which parts may happen in parallel.
Example: confirm permission and event facts; choose the invitation approach; approve any message before it is sent.
Check facts, dates, tone, privacy, accessibility, required approvals, and the actual task goal.
Use sample or confirmed details only. What did the workflow produce, and what did you review before using it?
Be specific. Example: the draft invitation still sounded like the venue was confirmed.
Change the instruction or checkpoint, then note how you would test that change next time.
Your notes save as you type and are included in the local learning-work export.
Test your playbook once, record one weakness and improvement, then mark the lesson complete when ready.