Show, Don't Just Tell

An agent flags a small thing: a tab in an internal tool reads "Guest Prof..." with no way to see the rest. It looks like a one-line fix. Truncate less aggressively, add a tooltip, ship it.

That is not what happened. The truncated label was a symptom. The real question was one nobody had actually answered: how should a row of tabs behave as more of them pile up. Nobody had defined that. So before any screen got touched, the problem got renamed, from "fix this label" to "define this behavior."

That renaming is most of what design work is. The screen is the last thing produced, not the first. Most of the effort goes into getting the problem right and making the tradeoffs visible enough that a team can choose between them. An artifact, a sketch, a clickable prototype, a diagram, exists to trigger that choice. It does not need to be polished. It needs to be real enough to react to.

AI does not change that job. It changes how fast you can get to the reacting part.

The loop, not the tool

Here is roughly how it went, using that tab bar as the example.

• Start from a real observation, not a spec. There was a screenshot and a hunch, nothing more. The first thing worth asking AI for was not "make this pretty." It was help comparing two or three plausible behaviors before committing to one.
• Make it tangible fast. The idea became a working, clickable prototype in minutes rather than days: a strip of tabs, a slider to shrink the available width, another to add more tabs. Reacting to something you can actually poke at beats arguing about something everyone is picturing differently in their head.
• Iterate on specific feedback. "It still truncates when there is room to spare." "The tab I am currently on should never get cut off." Each note went back into the prototype and it updated. See it, react, refine. That loop is the whole game, and it is the part AI shortens the most.
• Check it against reality. At some point the prototype got compared against the actual product requirements, and a gap turned up: the number of tabs a person might realistically have open had never been defined. Rather than guessing, that became a specific question to bring back to the team.
• Package it for a decision. The last step was not a finished design. It was a short rationale plus a scannable summary, built so a room of people could look at it and choose, not so it would look done.

Try it yourself

The prototype itself is the best explanation of what "make it tangible fast" meant in practice. Drag the width slider down, or push the tab count up, and watch three behaviors kick in, in order: full labels while there is room, icon-only tabs once space gets tight, then an overflow menu once even icons will not fit. The active tab is exempt the whole time. It never truncates and never gets pushed into the menu.

What AI actually did

It collapsed the distance between having an idea and seeing it, which meant more ideas got tried instead of one getting defended. It handled the parts that are tedious but not hard: measuring, checking the math, comparing the draft against the spec, drafting the question for the team. And it kept a record along the way, so the thinking stayed reusable instead of living only in someone's head.

What it did not do

It did not decide which problem was worth solving in the first place. It did not know what "good" feels like for someone actually doing this job under real conditions. It did not know when to stop exploring and go ask a person. And it did not make the final call. AI amplified the judgment already in the room. It did not supply judgment that was not already there.

That distinction matters more as these tools get better at producing convincing output. A prototype that looks finished is not the same as a decision that has been made well. The looking-finished part is now cheap. The deciding-well part is not, and was never going to be.

Where this leaves someone starting out

You do not need to draw beautifully to start doing this work. You need to be able to name a problem clearly and make it real enough for someone else to react to. A prototype is not a finished piece of work waiting to be admired. It is a question, given a shape.

Used well, AI lets someone newer to the field punch above their weight, not by producing more polish, but by trying more, learning faster, and spending saved time on the judgment calls that were always the actual job.