Part II: Building the first version
14. Test it, fix it, and keep going
Check the experience yourself and use the AI to fix specific problems.
Try the product from the beginning. Use a fresh request and type a response you haven't used before. Follow the steps your friend will take, including leaving the page and coming back to it.
You're checking whether the result matches the promise in your brief. The AI can help run checks and investigate failures, but you still need to see the experience you're about to give someone else.
Test a task, then its awkward moments
Start with the happy path from chapter 9. Can the designer create a request, can the client respond, and can the designer read the answer? Try the actual links the product provides, rather than opening a page through a shortcut only you know about.
Then try a few situations that could interrupt that task:
- Submit a form with something required left blank.
- Open a request as someone who shouldn't have access.
- Refresh after saving and check the result.
- Try a long question and response on a narrow screen.
Each check should answer a question about behaviour. If you submit an empty form, where does the product explain the problem? Does it keep the other information you already entered? Can you correct the field and continue?
Also check the first visit, before your own test data has filled the screens. A page that works well with several requests may leave a new designer with no indication of how to create the first one.
Describe the bug so it can be reproduced
When something goes wrong, record the starting point, what you did, what you expected, and what actually happened. That gives the AI a useful problem to investigate.
If a response disappears after a refresh, you could report it like this:
Starting point: a new feedback request opened through the client's
access path in a separate browser session.
I wrote a response and submitted it. The page showed confirmation.
After refreshing, the response was missing. The designer's view
also showed no response.
I expected the submitted response to remain saved and be readable
by the designer. Investigate where the save fails before changing
unrelated parts of the product. Fix the cause, run the relevant
checks, and tell me how to repeat this test.Use the same starting-point, action, expected-result, and actual-result structure for your own product. Include the error message if one appeared. If the AI asks for logs, ask where to find the application's record of errors and activity and which entries it needs. Keep passwords and secret keys out of anything you share. Describe what you know; leave the cause open until there's evidence for it.
Check the fix yourself
Repeat the same steps after the change. A successful result elsewhere in the product doesn't establish that this bug is fixed. Then try the nearby behaviour that the change could have affected.
Ask the AI to add a focused automated check when a failure is important or likely to recur. Saving a response to the correct request and preventing another user from reading it are useful behaviours to protect. A small wording change usually needs a visual check rather than a new test suite.
Ask the AI to run the checks the project already provides and explain any failures. A build check can find code that cannot be turned into a runnable application. A behaviour test can check a rule, such as whether a response belongs to the right request. Keep trying the product yourself too, including checking whether the instruction above a button makes sense to a new user.
Keep a working version within reach
Save a checkpoint after a meaningful piece works. If a later change makes the product worse, you can inspect what changed or return to that version and try another approach.
Code history and stored user data are different things. Returning to an earlier code version doesn't automatically undo changes made to the database. Before an experiment changes or deletes records, make sure you're using disposable data and understand how you'll recover what you need.
Decide what needs fixing before sharing
Fix anything that blocks the main task, loses work, or gives the wrong person access. Make confusing instructions clear enough that your friend can attempt the task without you operating the product for them.
Keep cosmetic improvements and additional features on a separate list. You can show a useful version with a plain layout. Record known limitations so you can explain them when you invite your friend, then take the working version into deployment.