Gather your files and start the interview
Download the blank template and save it in My Agent Files, or another folder you can find again. Karen's filled-in card from Session 4 shows what a finished one looks like. Read it on its own rather than attaching it to the interview.
YOUR RESOURCE
Reusable Workflow Card template
Download fileThe 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.
YOUR RESOURCE
Karen's example Reusable Workflow Card
Download fileKaren Wolfram's completed card from the Session 4 live session, a Post-Meeting Program Follow-Up workflow. Read it for the level of detail, then write your own.
Start a new chat in the Project/Notebook that holds your agent files, or attach your Agent Profile and Safety & Security Rules to a new chat. Attach the blank template too, or paste each file's complete text if you can't attach. Keep the whole interview in this one chat. If you haven't made the profile or rules yet, the two links at the start of this guide walk you through them.
Send the prompt below with no edits. The agent should confirm it has your three files, then ask whether you have a task in mind or would like help finding some.
Interview me for Reusable Workflow CardsYou have my Agent Profile, my Safety & Security Rules and a blank Reusable Workflow Card template in this conversation. If any of the three is missing, tell me which one and wait for it. Interview me to turn my recurring work into Reusable Workflow Cards. A card tells you how to handle one recurring process the same way each time. It gives you no permission to act on your own or to skip my review. How to interview me: - Ask one question at a time and wait for my answer. Never send me a list of questions. - I may type or use dictation. Pull the useful details out of rough answers, and keep my words where you can. - Don't re-ask anything my profile already answers. - Don't invent tasks, steps, sections, names, numbers, tools or standards. If I don't know something, write TO CONFIRM rather than guessing. - Follow my Safety & Security Rules throughout. Don't ask for or accept customer or employee records, credentials, or anything else my rules keep out. If an answer would need them, ask me to describe it in general terms or to use a fictional or non-sensitive version. - A card describes a process, so you never need the actual material to build one. If I'm not sure whether an input is permitted, write that as TO CONFIRM and carry on with the card. - If I ask to pause, stop, and pick up from the same question when I come back. Part 1, find the tasks. First ask whether I already have a task in mind or would like help finding some. - If I have one, screen it in Part 2. - If I'd like help, ask how my days and weeks actually run: my recurring meetings, reports, requests and updates, and what takes longer than it should. After about five questions, list up to five candidate tasks in my words and ask which ones to screen. I can add my own. Part 2, screen every candidate I chose, one at a time, before you build any card. Ask, one question at a time, what outcome I need from it, how often it comes up, and roughly how long it takes me today. Then tell me in two or three sentences whether AI is good at this work and whether it is worth a card. Don't estimate how much time a card would save. A task is worth a card when it repeats, follows a recognizable sequence and needs a consistent result, especially if I'm tired of explaining it. A one-off task, or a reply of a sentence or two, is not worth one. Recommend making a card or skipping it, give one reason, and let me decide. If I skip every candidate, offer to help me find more, as in Part 1, before you finish. Part 3, build each card. For each task I keep, start by proposing a workflow name and a one-sentence purpose from the screen, and let me correct them. Then ask about the six parts in order: Trigger (when it applies and when it doesn't), Input, Steps, Standard, Output and Review. For Standard, ask me what a good result looks like. If that is hard for me to describe, ask me to describe a past result I was happy with, or to share a fictional or non-sensitive example, and turn what makes it good into Standard lines I could check a result against. You can also show me the template's Standards to consider and add only the ones I pick. Before each new question, recap in one or two sentences how you would write my last answer into the card, so I can correct you. When a card's questions are done, fill in the template for it: - Keep the template's sections, headings and order, apart from the sample lists below. - Fill in the workflow name, its purpose and the six parts from my answers. - Keep the opening section, RULE PRIORITY, the rule lists in each part and HOW I WANT MY AI AGENT TO USE THIS WORKFLOW word for word. - Replace the bracketed placeholders with my content, using as many lines as my answers need. Delete surplus placeholder lines rather than padding them. If a section is left with none of my content, write one TO CONFIRM line there naming what it needs. Otherwise write TO CONFIRM only for a specific thing the card needs that I didn't know, never an open-ended line such as "any other checks". - Remove the three sample lists with their labels: Examples (delete these before you save the card), Example (delete this before you save the card) and Standards to consider. Copy a standard from that last list only if I picked it. - Give me the whole card as a downloadable plain-text file named Workflow-Card-SHORT-NAME-v1.txt, where SHORT-NAME is the workflow name in a few words joined by hyphens. If you can't create a file, put the whole card in one code block, with the filename on the line above it. After each card, ask whether to build the next task I kept or stop for today. When I stop, write a line that says NOT MADE INTO CARDS. Under it, list each candidate I skipped with my reason in one line, then each task I screened and kept but haven't built yet, marked build later. Leave out anything we never screened. Keep this list out of the cards. Start with your first question.
Choose your tasks and screen each one
Answer in your own words, as roughly as you like. Dictation works well for talking through an ordinary week. Describe each task by what it involves, never by the people or records inside it, and keep to the data rules.
If you have a task in mind, describe it in a sentence or two. If you built a card in Session 4, bring a different task here, and sharpen that card with the prompt in step 4 instead. If you'd like help, the agent asks how your days and weeks run, such as your recurring meetings, reports and requests and what takes longer than it should. Then it lists candidates in your words for you to choose from, and you can add your own. Good candidates often look like a weekly update you rebuild from scratch or a monthly budget check at home.
For each candidate, the agent asks what outcome you need, how often it comes up and roughly how long it takes you today. Then it says whether AI is good at the work and whether it's worth a card. A task earns a card when it repeats, follows a recognizable sequence and needs a consistent result, especially one you're tired of explaining. A one-off job, or a reply of a sentence or two, doesn't. AI is good at organizing, extracting, summarizing, comparing, drafting and reformatting, while approvals, commitments and decisions stay with you.
The agent recommends a card or a skip, and you decide, so say so if you disagree and give your reason. If none of your tasks pass, ask it to help you find others. If several pass, build one or two now and tell the agent when you want to stop.
Describe each task and what good looks like
For each task you keep, the agent asks about the six parts of a card, in order: Trigger (when it applies, and when it doesn't), Input, Steps, Standard (what a good result looks like), Output and Review (what you still check before it's used).
Before each new question, the agent recaps how it would write your last answer into the card. Correct it in the same chat if it adds a step, rule, name, number or tool you didn't give.
Show it what good looks like
A vague Standard is a common reason a result follows every step and still isn't usable. When the agent asks what a good result looks like, describe one you were happy with: its length, what came first, what it never left out. If that's hard to put into words, give it a fictional or non-sensitive version of a good result and let it pull out the qualities. Each Standard line should be something you could hold a result up against.
When a card's questions are done, the agent returns the complete card in the template's structure, as a file named like
Workflow-Card-Weekly-Status-Update-v1.txtor in one code block under that filename. Anything the card needs that you didn't know says TO CONFIRM. Then it asks whether to build the next task or stop. When you stop, it writes the NOT MADE INTO CARDS list: each skipped task with your reason, and each task you kept for later.
Check, save and try each card
Read each card against your answers. The Trigger should say when the workflow applies and when it doesn't, the Steps should follow your order, the Standard should hold what you said good looks like, and Review should name what you check before use. The rule lists under each part come from the template and stay as they are. Fill in each TO CONFIRM yourself or with the agent in the same chat.
Save each card as its own plain-text file in My Agent Files under the filename the agent gave, which ends in v1 for a first card. Keep the NOT MADE INTO CARDS list in a note beside them, for when your work changes.
Try one card in a fresh chat with your Agent Profile, Safety & Security Rules and the card attached, plus your Communication Style Guide if the result is something you send, or in your Project/Notebook with the card added. Send the prompt below with this run's material, from real work your company allows or a fictional or non-sensitive version. For a personal task such as a budget check, use made-up figures and keep to the data rules. If the agent stops to ask for something your card's Input part lists, attach it in the same chat.
Run a Workflow Card and check it against its StandardTask: Run my [workflow name from your card] workflow, using the Reusable Workflow Card in this chat. Context: [paste or attach this run's material: real work your company allows in this tool, or a fictional or non-sensitive version] Constraints: Use the workflow defaults. [add a one-time change, or delete this line] Output Check: Use the card's Output. After the result, write a line that says STANDARD CHECK. Under it, take each line of the card's Standard in turn. Say whether the result meets it, misses it or you can't tell, and quote the part of the result that shows it. For each miss, name the one part of the card (Trigger, Input, Steps, Standard, Output or Review) that most likely explains it, and suggest the smallest change to that part. Do not rewrite the card.
Below the result, a STANDARD CHECK marks each Standard line as met, missed or can't tell, quoting the result. Check the quotes yourself, because the agent marking its own work isn't proof. If a line was missed, change only the part of the card the check points to, save it as v2 and keep v1, replace the copy in your Project/Notebook if you use one, then rerun the same material in a fresh chat. If every line passes but you wouldn't use the result, find the part that should have prevented it: Standard for a missing quality, Output for the wrong shape, Steps for a skipped action. Then change it the same way.
Confirm your result
Confirm the outcomes below to finish this guide.
Final checks
A Reusable Workflow Card for a task that passed the screen is saved as its own file, and one run of it has been checked against its Standard.
Every task you screened either has a card or is on your NOT MADE INTO CARDS list with your reason.
Provide Feedback
Thanks. That went straight to the people who write these guides.