A customer opens a downloaded template and asks, “What do I do first?” If the answer requires a long support conversation, the product is not finished just because the file looks polished.
AI can help create guides, worksheets, examples, checklists, and small software tools. The work that makes them valuable is deciding what the user should accomplish, checking whether the product supports that task, and removing friction from the first few minutes of use.
Your work product is a small prototype with instructions, examples, a rights record, and an evaluation plan. Cedar Desk Studio's Job Log Starter Kit is a fictional teaching product. The included files are a draft prototype, not a released commercial product or a product tested with customers.
Define a finished task before choosing a format
For the proposed kit, the user should be able to record a request, assign responsibility, and identify the next action for waiting work. That is a more useful specification than “make a professional business template.”
Choose the simplest format that supports the task. A plain table may be enough for an individual. A shared system with reminders and permissions may be necessary for a team. Do not promise software behavior from a static file.
For this prototype, Cedar supplies a CSV file, a completed example, and a Markdown user guide. CSV stores a table as text and can be imported by many spreadsheet applications. The prototype contains no formulas, automatic reminders, account integration, or enforced workflow rules. Its written rules still require the user to follow them.
This boundary keeps the first version understandable. If research later shows that reminders are essential, Cedar can evaluate a different format and the additional costs of supporting it.
A workbook may add prompts and space for user notes; an interactive file may guide choices or validate entries. Test those features in the actual supported application and provide a usable static alternative where needed. Choose them because they help the task, not because a longer or more elaborate bundle looks more valuable.
Specify the information the task needs
The proposed job log uses eight fields:
| Field | Purpose |
|---|---|
| job_id | Unique reference for the request |
| opened_date | Date the request entered the log |
| summary | Short description of the work |
| status | new, in_progress, waiting, or done |
| owner | Person or role responsible for the next step |
| next_action | Concrete action still needed |
| next_review_date | Date to check open or waiting work |
| completed_date | Actual completion date when status is done |
Use unambiguous dates such as 2026-09-08. Explain blank fields: a missing owner means unassigned, not “the team.” A blank completion date means no completion date is recorded.
The prototype's proposed rule requires an owner, next action, and review date for in_progress and waiting rows. A done row requires a completion date. New unassigned work must remain visible for assignment. These are teaching rules, not an approved replacement for any business's operating procedures.
Do not collect sensitive data because there is space in the table. Use a case reference and a brief operational summary. Store any private customer material in the appropriate controlled system, not in every copy of a downloadable log.
Use AI to draft the product in layers
First create the specification, then the blank asset, then a completed example, and finally the instructions. Review each against the same task definition.
Create a draft Job Log Starter Kit from this specification.
Deliver a blank CSV header, three fictional example rows, and a concise
user guide explaining the first task and each status.
Use exactly the specified field names and ISO-style YYYY-MM-DD dates.
Label all sample data as fictional. Explain that the file does not
send reminders, enforce rules, or connect to other systems.
Preserve unknowns. Do not invent customer information, certifications,
test results, automatic features, or a commercial license decision.
List ambiguities and proposed evaluation tasks separately.
A useful example shows different states, not three copies of the same completed task. One row can be in progress, another waiting on information, and a third done with its completion date. The user can then compare an actual case with a relevant example.
Keep terminology stable. If the file uses next_review_date, the guide should not suddenly call it a guaranteed customer callback deadline. A review date is a date to inspect the work; any customer commitment needs its own agreement and evidence.
Make the first five minutes clear
The user guide should answer a practical sequence:
- What is this file for?
- What software and import settings does it need?
- Which file should I copy before entering my own data?
- How do I add one request and identify its next action?
- How do I find waiting work and record completion?
Include one worked walkthrough. For example: add J-001, describe the request, set an owner, choose in_progress, record the next action, and set a review date. When the action is blocked, change the status to waiting and state what information is missing. Mark it done only when the defined work is complete, then enter the actual completion date.
Explain how to preserve the sample. A blank master, an example, and a working copy serve different purposes. Someone should not have to delete fictional customer rows without knowing whether a formula or hidden rule depends on them.
Make troubleshooting concrete. If everything imports into one column, check the delimiter setting. If an identifier or date changes unexpectedly, review the import type and keep those fields as text when appropriate. Verify the actual software you plan to support before advertising compatibility.
Check function separately from presentation
Open or parse the delivered files and check their structure. Are the fields consistent? Are example IDs unique? Are dates valid and in sensible order? Does every done example have a completion date? Does waiting work show an owner and next action?
The downloadable practice script, included with its fixtures in the Part 6 practice pack, checks these properties for the included examples and deliberately invalid practice rows. That is a file-consistency check. It does not prove that customers can use the kit, that a spreadsheet application imports it correctly, or that the file enforces the written rules.
Then plan an observed task session with intended users. Give a realistic task and watch what happens before teaching the solution. GDS's moderated usability testing guidance describes observing participants attempting tasks to understand usability problems. Apply the method to the actual audience and product.
Record where people hesitate, misinterpret a field, lose data, or need help. “They liked the colors” does not answer whether they can find overdue waiting work. Keep actual observations separate from a model's imagined user reaction.
Make the delivery and instructions accessible
Use readable text, meaningful headings, descriptive links, and clear labels. Do not rely only on color to distinguish waiting from done. Provide text explanations for visual examples and review alternative formats when the chosen file creates a barrier.
Check the product's landing page and delivery flow as well as the file. A keyboard user should be able to reach the download and help route. W3C's Easy Checks provides an initial web accessibility review, while explaining that a full evaluation requires more than those preliminary checks.
Do not claim that every PDF, spreadsheet, or web page is accessible because it opens. Evaluate the actual format with relevant assistive technology and users. Record which checks were performed and which remain outstanding.
Keep a rights and authorship record
List where the text, illustrations, data, fonts, and code came from. Keep the relevant license or permission and note whether it permits the particular use: editing, commercial distribution, inclusion in a template, or redistribution of an original asset.
Being allowed to use an image in your marketing does not necessarily mean you may resell the image file inside a template kit. A supplier's permission to list a product also does not automatically cover unrelated digital products. Check the actual terms rather than asking the model for a reassuring general answer.
In the United States, the Copyright Office explains that AI-assisted work may contain protectable human authorship, while merely providing prompts does not by itself supply sufficient expressive authorship for generated output. Protection depends on the actual human contribution. See the Office's copyrightability report announcement.
Keep that question separate from permission to use third-party material and from a tool provider's contract. A commercial-use term does not establish that every generated output is exclusive, non-infringing, or wholly copyrightable. Record human contributions honestly and review rights questions that matter to the product before selling it.
For this teaching prototype, the records identify AI-assisted preparation and pending owner review. They do not invent interviews, a human design team's contributions, or a completed commercial licensing decision.
Package the product around the buyer's task
Give the folder a clear entry point: a README or user guide that names the current version and explains each file. Include the blank asset, example, limitations, and actual support arrangements. Keep internal test cases outside the customer's normal working file.
The product page should accurately state formats, prerequisites, included files, and supported tasks. If the download is a static CSV, do not illustrate a dashboard with automatic notifications unless that functionality is actually included and verified.
Budget for the whole product: source review, editing, file checks, accessibility, delivery testing, support, and maintenance. Faster drafting can make a small product economical, but a large untested bundle can create more support work than it saves.
Finish with a release decision that distinguishes a draft prototype, a technically checked asset, a user-evaluated product, and an approved commercial release. The included Job Log Starter Kit reaches the draft and file-check stage only.
Next: Plan how the product will be supported, updated, and eventually retired in Chapter 25.