Ready to put this into action?
Get the complete AI Integration Playbook β Practical AI implementation guide β prompt engineering, workflow automation, and ROI frameworks.
Article 020 Β· Part 2
Build a Small Library of Reusable Workflows
Save the instructions, inputs, and checks that make a recurring task work.
By Randy Salars Β· Published
On this page
- Choose tasks with a recognizable shape
- Card 1: Turn confirmed facts into an announcement
- Card 2: Turn meeting notes into decisions and actions
- Card 3: Describe an item from its record
- Card 4: Run a study session that requires an attempt
- Card 5: Compare options using explicit criteria
- Keep instructions separate from examples and live data
- Test all five on fresh inputs
- Then test missing and conflicting information
- Maintain a useful collection
Save the instructions, inputs, and checks that make a recurring task work.
Every week, you ask for the same kind of help. Turn meeting notes into actions. Draft an announcement. Describe an item. Prepare a study session.
Every week, you remember another instruction after the first answer arrives: keep the dates, do not invent an owner, wait for my attempt, flag missing prices.
A small collection of workflow cards can save that repeated explanation. Each card describes a task, the information it needs, the output it should produce, and how you will check the result.
You do not need a special app to begin. A document or folder with five clearly named cards is enough. The value comes from the working process you preserve.
Choose tasks with a recognizable shape
A reusable workflow works best when the task repeats with new inputs. The event changes, but the announcement still needs a date, place, purpose, and next action. The meeting changes, but action items still need an action, owner, deadline, and status.
Start with tasks you actually perform. Do not collect a hundred elaborate prompts for work you may never do.
For each candidate, ask: Are the required inputs predictable? Can I inspect the output? Do I know the common mistakes? If the task is still poorly understood, work through a real example before turning it into a template.
These cards guide work you review. They do not create an autonomous system, connect an application, or authorize external actions simply by being saved.
Card 1: Turn confirmed facts into an announcement
Use when: You have current event details and need a draft for a specific audience.
Required inputs: Audience, confirmed facts, desired response, length or format, and any explicitly unresolved details.
Copyable task brief:
Draft an announcement for the audience below using only the confirmed facts. Put the event's purpose and essential attendance details where readers can find them. End with the supplied next action. Preserve conditions such as optional participation or unconfirmed arrangements. If a required fact is missing or contradictory, flag it separately; do not invent it. Follow the requested length and format. Return the draft and a short review list of unresolved items. Drafting only.
Audience: [supply audience]
Confirmed facts and source date: [supply current facts]
Next action: [supply response or attendance instruction]
Format and length: [supply requirement]
Unresolved details: [list them or state none known]
Review: Compare every date, time, place, number, condition, and promise with the inputs. Confirm the message is ready before distributing it.
Card 2: Turn meeting notes into decisions and actions
Use when: Notes contain a mixture of discussion, decisions, and work people agreed to do.
Required inputs: Notes you are permitted to use, meeting date, and any corrections supplied by participants.
Copyable task brief:
Organize these meeting notes into three sections: confirmed decisions, assigned actions, and unresolved questions. For actions, use columns for action, owner, deadline, and supporting note. Preserve tentative language. If an owner or deadline is absent, write βnot assignedβ or βnot specified.β Do not turn suggestions into decisions or infer a person's agreement. Flag contradictions for review. Use only these notes and corrections.
Meeting date: [supply date]
Notes: [supply permitted notes]
Participant corrections: [supply corrections or state none]
Review: Check that each decision was actually made and each owner actually assigned. Ask participants to resolve ambiguous notes through your normal process.
Card 3: Describe an item from its record
Use when: You need consistent descriptions of inventory, collection objects, or items for a sale.
Required inputs: One record per item, factual description fields, condition, and price if applicable.
Copyable task brief:
Create one short description per item. Start with its supplied name and price; if price is absent, write βprice not supplied.β Follow with one or two plain sentences covering supplied features and condition. Keep defects explicit. Use only the individual item's record. Do not transfer facts between items or add claims about rarity, safety, authenticity, or performance that the record does not establish. List conflicting fields for review. Follow the labeled structural example if one is supplied.
Item records: [supply records]
Optional example, for structure only: [supply an authorized example or omit]
Review: Match every description to its own record. Confirm missing information remains missing and limited observations have not become guarantees.
Card 4: Run a study session that requires an attempt
Use when: A learner wants practice on a defined skill using permitted assistance.
Required inputs: Topic, current understanding, approved reference material, and class rules where applicable.
Copyable task brief:
Help me practice the skill below within the supplied rules. Use the approved reference material for factual explanations. Give one new practice question at a time and wait for my attempt. If I struggle, offer a small hint before a complete worked solution. Identify the first meaningful error and ask me to try that step again. Do not claim an answer is correct without checking it. At the end, ask me to explain the skill and attempt a fresh question without help. Flag any gap or contradiction in the reference material.
Skill and current understanding: [supply details]
Reference material: [supply permitted material]
Assistance rules: [supply applicable rules]
Review: Check explanations against the reference or another suitable method. Judge progress through independent attempts, not through how easy the explanation felt to read.
Card 5: Compare options using explicit criteria
Use when: You need a decision aid based on supplied facts.
Required inputs: Options, current evidence, constraints, and the criteria that matter to you.
Copyable task brief:
Compare these options against the listed criteria and constraints. Keep supplied facts, assumptions, and judgments distinct. Use a table where it helps comparison. Mark unknown information rather than inventing it. Identify any option that fails a stated hard constraint. Explain the main tradeoff and the missing fact most likely to change the choice. Do not create numerical scores or weights unless I supply or request them. End with the next useful decision or verification step.
Options and source information: [supply material]
Hard constraints: [supply limits]
Evaluation criteria: [supply priorities]
Review: Verify the figures and ask whether the recommended option still makes sense if an important assumption changes. Your priorities determine the decision.
Keep instructions separate from examples and live data
The bracketed fields are places to supply new information. They are not facts for the assistant to infer.
Keep each saved card free of private records. Store any demonstration as a clearly labeled fictional or authorized example. Put the current task's actual inputs in a separate section when you use the card.
This avoids a common failure: a template created for one event keeps inserting that event's date into the next announcement.
If you adapt a card for a team, name the person who maintains it. Shared instructions need someone who can resolve conflicting edits and remove stale material.
Test all five on fresh inputs
Try each card with a new input unrelated to the conversation where you created it. Here are small fictional cases and their acceptance checks:
| Card | Fresh practice input | What to check |
|---|---|---|
| Announcement | Drawing club, November 21, 2026, 1β2 p.m., Oak Room, free; bring a pencil; reply if interested | Every detail survives; no materials or refreshments are promised |
| Meeting notes | βLena agreed to bring labels by November 19. A larger sign was suggested; no decision.β | One assigned action with its deadline; sign remains a suggestion or open question |
| Item description | Wooden tray, 12 inches long, scratch on underside, $5 | Size, scratch, and price preserved; no unsupported wood species |
| Study session | Practice adding whole numbers; supplied reference: 14 + 9 = 23, because 14 + 6 + 3 = 20 + 3 | One new question; wait for an attempt; check arithmetic; final independent problem |
| Option comparison | Room A: $40, 12 seats; Room B: $70, 20 seats. Need 18 seats; budget $80 | A fails capacity; B fits supplied cost and capacity, with other suitability questions still open |
These are proposed test inputs and expected checks, not a report that a particular model passed them. Save your actual outputs and observations.
Then test missing and conflicting information
Use the announcement card for two harder trials. First remove the event time. The result should identify the missing time; a provisional draft must not invent one or be presented as ready to send.
Next supply two equally authoritative records with different event dates. The assistant should flag the conflict and ask which governs. Choosing the later-looking date without justification is a failure.
You now have three trials for that card: ordinary, incomplete, and conflicting input. If it fails, change the instruction that addresses the observed problem and retry the affected case. Do not add twenty unrelated restrictions.
Google's prompting guidance explicitly treats refinement as iterative and dependent on the task and observed response. Apply that principle to your own cards instead of assuming a downloaded template has been validated for your work. Google: Prompt design strategies.
Maintain a useful collection
For each card, record its purpose, version, owner, last actual test date, and known limitations. A simple note might say, βVersion 2: now flags conflicting event dates; tested on three cases on the recorded date.β Fill that date only after you run the tests.
Review a card when a real failure appears, the source rules change, or the task changes. Retire cards that duplicate others or cost more effort than doing the work directly.
For students, a saved learning card should change as understanding grows. Move from hints to independent problems when you are ready. A workflow that always provides the next step may eventually hold you back.
Completion check: You have five usable cards, have tried each on fresh information, and have tested one with ordinary, missing, and conflicting inputs. You can show the outputs and explain which checks passed or failed.
Your growing skill is visible in those decisions: you can describe work, review it, preserve what matters, and carry a successful process into the next task.
Get the AI Dispatch
Weekly insights on ai & technology β delivered to your inbox. No spam, unsubscribe any time.
Want to choose specific topics? Customize your interests