Define the user and useful action
Describe who the tool serves, the input it accepts, the decision it supports, and the result it returns.
✦ Course · Intermediate
Build a live artifact or custom AI tool
Turn a business need into an interactive artifact or customer-facing tool, test it in preview, and prepare it for safe delivery.
Start courseYOUR OUTCOME
You will create and test a bounded interactive tool for a real user need.
Define user and decision Build in the correct surface Test behavior and data boundaries
6 min
Choose artifact, custom tool, or code project
Use an interactive single-purpose tool for a focused workflow and Code agent for multi-file applications.
Provide rules and sample data
Specify calculations, validation, states, copy, brand treatment, and realistic examples without exposing sensitive data.
Generate and steer the build
Review the live preview, correct assumptions early, and keep accepted behavior intact.
Test the edges
Check empty, invalid, unusual, mobile, permission, and failure states before considering the tool ready.
Export or publish deliberately
Review included files, dependencies, destinations, access, and customer data handling before delivery.
What each lesson covers
The course, lesson by lesson.
Define the user and useful action
Describe who the tool serves, the input it accepts, the decision it supports, and the result it returns.
Before building anything, type: I want a tool for our sales reps. Input: a customer's monthly energy use in kWh and their current tariff. Decision it supports: which of our three plans saves them most. Output: the yearly saving and a one-line recommendation they can read aloud. Confirm this back before you build.
Choose artifact, custom tool, or code project
Use an interactive single-purpose tool for a focused workflow and Code agent for multi-file applications.
Type: Build a Custom Tool for HR: staff enter their start date and the leave they have already taken, and it shows their annual leave balance. One screen, no login, no database.
Provide rules and sample data
Specify calculations, validation, states, copy, brand treatment, and realistic examples without exposing sensitive data.
For a landscaping quote calculator, type: Rules: labour is 85 dollars an hour, materials carry a 15% markup, and jobs over 5,000 dollars must show a 30% deposit. Validate that hours and materials are positive numbers and show a clear message if not. Use these three sample jobs with invented client names, not our customer list.
Generate and steer the build
Review the live preview, correct assumptions early, and keep accepted behavior intact.
With the event-registration tool open in preview, submit a test registration, then type: The confirmation message is right, keep it. But choosing Student in the ticket-type dropdown does not show a student ID field. Add that field and make it required only for Student.
Test the edges
Check empty, invalid, unusual, mobile, permission, and failure states before considering the tool ready.
With the class-fee calculator preview open, type: Run these cases and tell me what happens in each: no fields filled, a negative number of students, 10,000 students, a phone-width screen, and the price list failing to load. Fix anything that breaks and list what you changed.
Export or publish deliberately
Review included files, dependencies, destinations, access, and customer data handling before delivery.
Before sharing the supplier-scorecard tool with the procurement team, type: List every file and external dependency in this tool, where it would be deployed, who could open it, and whether any supplier data is stored anywhere. Do not publish yet.
Keep going
Make the learning stick
Start with one real piece of work.
Open Oppermind beside the lesson, apply each step, and leave with something you can use.