Week 4 Companion

Build Your Reusable Workflow Card

Week 4Session Companion

  • 30–45 minGuided walkthrough

This guide is a direct companion to the Week 4 livestream. Follow along while you watch the live session or use this as a reference if you get stuck or miss some of the session and need to catch up.

What You Need for This Guide

  • Your , Communication Style Guide and Safety & Security Rules from Weeks 1 to 3, if you have them.
  • One piece of work you do again and again, and a new example of it that you’re allowed to use with your AI tool, such as this week’s notes for a status update. Public or made-up material works too.
  • Your approved AI tool and a plain-text editor.

Create your Agent Profile

Build your Communication Style Guide

Build your Safety & Security Rules

Choose the tool you’re allowed to use

Your approved tool

The tool your company approved.

Pick your approved tool above to see the steps and screenshots for it.

Start guide
5 steps · At your own pace
About saved progress

No account needed. Checkmarks save on this browser and device when browser storage is available. If storage is blocked or full, new checkmarks last until you reload or close this page. Progress doesn’t sync to other browsers or devices. Clearing this site’s browser data removes it, and private browsing may not keep it after you close the window.

Step

Choose one recurring process

In Week 1 you built your , so your agent understands your work. In Week 2 your Communication Style Guide taught it to communicate like you, and in Week 3 your Safety & Security Rules set how it handles your information. Those rules come with you from here on. Company policy and your Safety & Security Rules still govern, and nothing you build this week weakens either.

Your agent can know a lot about your work and still have no idea how you do a recurring piece of it: the order you work in, what you check first, which details matter, where missing information stops you, what a good result looks like, the format people expect and what you review before anything goes anywhere. When that process lives only in your head, the agent improvises, and even when a result happens to match how you work, nothing makes it happen the same way next week. Repeatable work needs a repeatable process. This week you’ll write the important parts of one process into a Reusable Workflow Card, a file your agent follows when you give it that kind of work. Capture the parts that decide whether the work comes out right, not every click. By the end you should be able to say of your agent, “It follows my process”: not perfectly, and not without your review, but consistently enough that you stop explaining the whole process from scratch every time.

The session separates three things. The task is what needs doing now, such as “create this week’s project status update”. The process is how that kind of work should be done every time it comes up. The result is why the business cares, such as leadership seeing risks and decisions sooner. This week is about the process, and Week 5 takes on the result. A Workflow Card also isn’t automation: it doesn’t connect systems, watch for work or let the agent act on its own. When you give your agent the work, the card tells it how you want that work done.

Now choose one recurring process you’d like your agent to follow more consistently. Look for work that repeats, stays fairly stable, follows a recognizable sequence and needs a consistent result, especially work you’re tired of explaining again. The recurring work you listed in your Agent Profile is a good place to start. For example:

  • A weekly status update
  • Turning meeting notes into actions
  • A recurring briefing
  • Preparing an agenda
  • Turning research into a recommendation
  • Reviewing a document

For a first card, pick something you know well and can test today, rather than the most complicated process you have or one where every case is completely different. If the real work needs information you’re not allowed to use with your AI tool, build the card around a made-up, public or sanitized version, because the skill you’re learning is the workflow. In the live session, keep your choice private and don’t post work details in the chat.

Step

Learn the six parts and where each instruction belongs

Since Week 1 you’ve framed each task with the Context Sandwich: Task (what you need done now), Context (the situation and source for this run), Constraints (the limits the result must respect) and Output Check (what the result must contain and what you’ll review). A Workflow Card doesn’t retire it, but it makes it smaller. If it stays the same every time, make it reusable. If it changes this time, keep it current. The card holds the defaults that repeat, and today’s sandwich holds what’s new: this run’s facts and source, a one-time exception or an extra piece of output.

Each instruction needs one best home, at the highest level where it’s reusable. Policy, legal, privacy and security rules go with your organization’s rules and your Safety & Security Rules. Background about your work goes in your , and how you communicate goes in your Communication Style Guide. A rule for this recurring process goes in your Workflow Card, and anything true only this time goes in today’s Context Sandwich. Two copies of one rule eventually disagree, so the card doesn’t restate your Safety & Security Rules: it tells the agent to follow them throughout.

A Workflow Card has six parts, and each one answers a different question:

  • Trigger: When does this workflow apply? Too broad, and the agent uses it where it doesn’t belong. Too narrow, and you end up with many nearly identical cards.
  • Input: What must be available before it can succeed? The card names the kinds of input, and each run supplies the actual ones.
  • Steps: What should happen, and in what order? Use verbs you could check afterward, such as identify, extract, compare, separate, check, draft and flag. “Analyze the notes” leaves the agent to invent the process between the lines.
  • Standard: What does a good result look like? This is the quality bar. “Return five bullets” is Output, and “those bullets separate risks from decisions and make the decision needed obvious” is Standard.
  • Output: What should it return? A predictable structure, so this week’s result has the same shape as last week’s.
  • Review: What must a person still check before the work is used? AI output is not approval, and when review is required, the workflow isn’t finished until it happens.

When instructions conflict, Rule Priority decides which one wins. Your Safety & Security Rules set the full order: company policy first, then those rules, then today’s task and Context Sandwich, then your Communication Style Guide and your Agent Profile. The card’s own Rule Priority line fits it in after today’s task and Context Sandwich and before your style guide and profile. Today’s sandwich can change one of the card’s defaults for a single run but never weaken policy or your safety rules, and the card tells your agent to point out a conflict that matters rather than choose silently. The two lists answer different questions: one best home says where to write an instruction, and Rule Priority says which one wins when two disagree. So today’s Context Sandwich is the last place you’d store a rule, yet it still outranks the card for the run in front of you.

Step

Build your first usable Workflow Card

Now build the card for the process you chose. Aim for a first usable version rather than a perfect one, because testing in steps 4 and 5 will show you where it needs work. As you go, notice what you usually repeat in your Context Sandwich for this work: the same task, audience expectations, constraints, output sections or checks. Those are the defaults to move into the card.

  1. Download the template, open it in a plain-text editor and save it in My Agent Files, or another folder you’ll find again, under your workflow’s name, such as Weekly Project Status Update Workflow Card. If it opens as a preview in your browser, copy its text into a new plain-text document and save that. Leave its opening, its Rule Priority and its closing section, How I Want My AI Agent to Use This Workflow, as they are. In the six parts, write over each line in square brackets as you go.

    YOUR RESOURCE

    Reusable Workflow Card template

    The blank Reusable Workflow Card Raymond uses in Session 4. Fill in its six parts, Trigger, Input, Steps, Standard, Output and Review, for one process you repeat, and save it with your other agent files.

    Download file

    For the level of detail, you can also download Karen Wolfram’s cards. Raymond shares four of her completed cards for this session, and together they show one person’s set of workflows: a follow-up record after a customer or program meeting, a readiness checklist for a training cohort, a stakeholder message drafted from that work, and a status and decision brief for leadership. Open the one closest to your own work and read it beside your card without pasting it into a chat. Her steps, reviewers and organization are hers, so write your own.

    YOUR RESOURCE

    Karen's example card: Post-Meeting Program Follow-Up

    Karen Wolfram's card for turning permitted meeting notes into a follow-up record of decisions, next steps, risks and open questions.

    Download file

    YOUR RESOURCE

    Karen's example card: Program Delivery Readiness Review

    Karen's card for keeping one 16-item checklist of what a training cohort needs before delivery, updated as permitted evidence arrives.

    Download file

    YOUR RESOURCE

    Karen's example card: Stakeholder Communication

    Karen's card for drafting a short message to a program stakeholder, internal or customer, ready for her review before anything is sent.

    Download file

    YOUR RESOURCE

    Karen's example card: Program Leadership Status and Decision Brief

    Karen's card for a short internal brief that shows leadership progress, milestones, risks and the decisions they need to make.

    Download file
  2. Under Workflow Name, give the workflow a name that says what it does, such as Weekly Project Status Update rather than My AI Workflow, and a sentence or two on its purpose. Under Trigger, finish “Use this workflow when…” clearly enough that you could tell whether it applies, and list a case or two under Do Not Use This Workflow When. The name and Trigger now carry the recurring job, so your Task line can shrink to “Run my Weekly Project Status Update Workflow.”

  3. Under Input, list the kinds of input the workflow needs and mark any optional ones: current notes, the reporting period, the audience if it changes from run to run, or a previous approved version if you compare against it. If the audience is always the same, leave it out of Input and write what they need into your Steps, Standard or Output, such as the sections they always read first. Name the kind of input, never the content itself. Asking for an input doesn’t make it permitted, and your Safety & Security Rules still govern what you give the agent. The template’s Input Rules already tell the agent to ask rather than invent when something important is missing.

  4. Under Steps, write the major moves in order, as if you were explaining them to a capable colleague who has never done this work: where to start, what to look for, what to separate, where to stop if something important is missing and where a person has to decide. Use verbs you can check afterward. A constraint you repeat every time can become a step, such as “If an owner isn’t named, mark it Owner Needed.” The agent can organize, extract, draft and suggest, but supporting a judgment isn’t owning it. When the process reaches an approval, a commitment or another decision that’s yours, write a step that stops there and makes the checkpoint visible. Three steps can be enough, and so can twelve.

  5. Under Standard, add two to five things that would make you reject the result if they were wrong, such as “tentative items stay visibly tentative”, copying any that fit from the template’s Standards to consider. A constraint you put in your Context Sandwich every time is a good candidate. A limit for one run stays in that run’s Constraints.

  6. Under Output, write the structure you want back every time, such as named sections or a table. If the shape depends on the audience, say so. Once the sections are here, your Output Check only needs what a run adds or changes.

  7. Under Review, list what you or another required reviewer checks before the work is used, such as names, dates, owners, commitments, audience fit and any approval. Phrase it as “Flag this output for my review before use”, because the agent can’t grant approval. A check you make on all your AI work belongs in your Safety & Security Rules, and an extra approver for one run goes in that run’s Output Check.

  8. First-version checkpoint: stop and scan the whole card. Can you answer all six questions: when do I use it, what does it need, what does it do, what does good look like, what does it return and what still needs a person? Then ask what can now disappear from your usual Context Sandwich because the card owns it. A rough field can stay rough until testing shows what it needs. Delete any line still in square brackets, and the examples and list of standards the template marks for deleting, so your agent reads only your process. Karen’s cards show what that looks like. Save the file.

Step

Run your card on a new example

Now test the card the way Raymond does in the session, with a deliberately lean Context Sandwich, so you can see whether the card carries the process by itself. Use a new example of the same work, such as another week’s notes or another meeting, different from the one you pictured while building. Same process, different content: that’s the test of whether your card is reusable.

  1. Choose your new example and prepare it the way the data rules and your Safety & Security Rules say, leaving words such as “proposed” and “pending” exactly as they are. If you don’t have a second example you may use, write a short made-up one.

  2. Before you run it, predict what should come back if the card works: your card’s output sections, the facts organized the way your steps say, owners and dates only where your source gives them, missing information marked rather than filled in, and the review checkpoint named.

  3. Start a new chat and copy Run my Reusable Workflow Card. Attach your Workflow Card, Safety & Security Rules, and Communication Style Guide, or paste their text under My Files, and write “not provided” for any file you don’t have yet. If the message is too long, attach the files rather than pasting them. If they still won’t all fit, send your Workflow Card and Safety & Security Rules, and write “not provided” for the others. Under Task, write your workflow’s name. Under Context, say what this run covers, such as the reporting period, and the audience if it’s different this time. Add a one-time constraint or an extra piece of output only if this run needs one, and otherwise leave the workflow defaults. Paste your prepared source at the end and send it.

    Run my Reusable Workflow Card
    Use the files I have attached or pasted below for this conversation. My Reusable Workflow Card describes how I want this recurring work done and how to use it with my other files. If my Workflow Card is missing, ask me for it before you start. Continue without any other file I've marked "not provided".
    
    MY FILES
    [paste the text of your Reusable Workflow Card, Safety & Security Rules, Agent Profile and Communication Style Guide here, or attach the files and write their names here. Write "not provided" for a file you don't have yet.]
    
    TASK
    Run my [name of your workflow].
    [add anything different about this run, such as a second version for another audience, or delete this line]
    
    CONTEXT
    [what this run covers, such as the reporting period or which meeting the notes are from, and the audience if it's different this time]
    
    CONSTRAINTS
    Use the workflow defaults.
    [add a limit for this run only, such as a shorter summary, or delete this line]
    
    OUTPUT CHECK
    Use the workflow defaults.
    [add anything extra this run needs, such as a decision to flag before a deadline, or delete this line]
    
    SOURCE
    [paste this run's prepared source here]
  4. Your agent may ask you something before it gives you a result, and that’s often your card working. If it asks for an input your card lists, answer in the same chat, or say you don’t have it so it stays marked as missing. If it stops at a decision your Steps leave to you, give your decision in the same chat, or tell it the decision is yours to make later and ask it to finish the rest. If it points out a conflict between your card and this run, say which one to follow.

Step

Check the result and fix the smallest thing

Don’t judge the result by whether it looks good, runs longer or has more formatting. Check whether the agent followed the process you gave it. With your source open beside it, go through the six parts:

  • Trigger: Did this work fit the workflow?
  • Input: Did the agent have what the card asks for, and did it ask about anything missing instead of inventing it?
  • Steps: Can you see each step in the result, such as decisions kept apart from next steps?
  • Standard: Is every important claim in your source, with qualifiers kept and no owner or date invented? If you added a one-time constraint, did the result follow it without the agent treating it as a change to your card?
  • Output: Is it the structure your card asks for?
  • Review: Does it name what you still need to check?

Also check that nothing from the example you pictured while building leaked in, and that your Safety & Security Rules were followed.

  1. If something’s wrong, diagnose it before you add more rules, and change the smallest part of the card that explains the failure:

    • Wrong situation: fix the Trigger.
    • Missing information: fix the Input.
    • Missing or vague actions: fix the Steps.
    • Steps followed, but poor quality: fix the Standard.
    • Right content, wrong shape: fix the Output.
    • Human checkpoint unclear: fix Review.
    • The same rule in two places, now disagreeing: keep it in its best home and delete the copy.

    If every part checked out, there’s nothing to fix.

  2. Make any change in your saved card, because a chat doesn’t change your file. Then start a new chat, send Run my Reusable Workflow Card again with your updated card, your other files and the same source, and check the six parts again. A longer card isn’t automatically a better one: useful beats bloated, and reusable beats repeated. If you notice you’re still typing the same instruction into your Context Sandwich every time, move it into the card, or into the file where it belongs.

  3. Save the card in My Agent Files. You can also copy Run my Reusable Workflow Card into a plain-text file there with your workflow’s name filled in, and edit it to fit how you run this work. Next time the work comes up, start a new chat with your files, replace the old details with the new run’s and check the result against the six parts again.

This week’s upgrade, Let Your Agent Interview You for Workflow Cards, has your agent interview you about the rest of your recurring work, check which tasks are worth a card and fill one in for each. Next week you’ll build your Business Results Workflow, moving from how recurring work gets done to the business result it improves.

Build Your Reusable Workflow Card

Confirm your result

Confirm the outcomes below to finish this guide.

Final checks

You have a saved Reusable Workflow Card for one recurring process, with its six parts filled in well enough to use.

You have a result from a new example, run with a lean Context Sandwich, and you can point to where it followed your process and what still needs a person’s review.

Provide Feedback

Thanks. That went straight to the people who write these guides.