Let the answer shape the form.
The next iteration gives the AI a richer interface vocabulary while keeping the person in control of what it understood.
The build journal / 2025 to 2026
Inside the work.
From the first request to the detail that changed the product.
Follow a project, choose a date, or find your own starting point.
Signing up one person did not test a shared home.
I asked for a household of four, then three separate people joining it.
Two dates. Two taps.
A vacation range did not need separate From and Until interactions before the calendar could do its job.
Three people at home should not mean nine days of waiting.
Adapting a chore’s rhythm needed a limit that people could understand.
Let the answer change the interview. Keep the question still.
An adaptive form still needs a stable place to read, answer, and understand progress.
One hat. Ten visual directions.
The product stayed the brief’s anchor. The light, setting, and role of the image changed.
Monthly should mean a month after it was last done.
Changing a chore’s schedule exposed an old assignment that should no longer have existed.
Making the navigation smaller made the collision worse.
The chat composer and the tab bar were competing for the same space.
The words are part of the interface.
I wanted English and German copy that helped restaurant staff act, not decode the software.
“Backend” was not a baking service.
A long software development description exposed both an input limit and a cross language false match.
A price needs a unit. A profile needs evidence.
A polished generated profile must not turn missing information into a confident claim.
The model answered. The app threw it away.
A response shape mismatch quietly turned a live AI interview into hardcoded questions.
The dishwasher is a sequence, not a checkbox.
Running it and emptying it are different jobs. The product needed to connect them.
A week of availability, painted in a few gestures.
The app had already filled in my schedule. Editing it was the hard part.
A reservation rule has to survive the final click.
A fast availability check is only one part of a dependable booking flow.
Zero problems should not need a dashboard section.
There were no duplicate guests to review. The interface still gave the restaurant owner a paragraph about it.
Could this actually run a hundred restaurants?
I asked a capacity question. The useful answer was a map of the order path’s weak points.
The design system had to reach the actual screens.
A blueprint and an implementation can both exist while still speaking different visual languages.
A portfolio is part of the answer.
Different services need different ways to show what makes them worth choosing.
Let me order dinner before you ask me to pay.
The checkout integration was working its way into the wrong moment. I asked the AI assistant to follow the restaurant journey instead.
Saved. But which part of the website?
One editing session, several different saving rules. How I worked with an AI coding assistant on a restaurant website builder, and why the smallest status message needed a much closer look.
Five QR designs later, I rewrote the brief.
The AI produced five visual directions. I rejected the result because I needed something that belonged on a restaurant table.