X.AI at work · July 2026
From a rough idea to a PR in under two hours, because I could prototype in the real thing.
A colleague from another team came to me with an idea of data they wanted to surface. Two hours later, working directly in the real repo with Claude beside me, we had a PR — and, more importantly, a much clearer idea of what they actually wanted to see.
By Gisella Famà · 4 min read · AI at work
Someone had an idea of what information we needed to surface in a table. The kind of idea when you half know the data exists, and half know what story you want it to tell.
Two hours later he had a pull request open, a much better representation, and — the part I care about most — a sharper opinion about what information they actually needed on the screen. None of that would have happened at the same speed if I had opened Figma.
"The fastest way to help someone figure out what they want is to put a working version of it in front of them, in the real product."
What actually happened
They showed me some data in Excel. I did the quick designer pass in my head — signpost the columns and rows, highlight the ones I couldn't recognise, add a header that says what this table is for. Nothing clever yet.
Instead of mocking it, I opened the repo. I asked Claude to pull the fields into the shape I wanted, wire them into the existing table component, and give me something I could load locally.
First shot got me to about 60%. The columns were right, the layout was close, some fields were still noisy. Good enough to react to, which is the point.
I tested it locally, made a handful of adjustments — mainly working on the content (from payload).
I showed him the running page, not a screenshot. He immediately started saying things like "actually I don't need this, but I do need this one" — which is exactly the kind of feedback you never get from a static mock.
A couple more passes, a couple more Claude questions, and I'd say we're at 85% of what good looks like. Not perfect. Definitely enough to open a PR and let the engineers finish the last stretch properly.
Why prototyping in the real environment changes the conversation
If I had brought him a Figma frame, we would have talked about the frame. He would have squinted, said "yeah, looks fine", and both of us would have carried a small quiet worry that once it was built we'd find out it wasn't what he meant. Instead he was clicking on a real page, with real data, in the real app. His feedback stopped being polite and started being specific.
That is the shift. The prototype is not there to impress him. It is there to help him think. A working page in his hands is a much better prompt than any question I could have asked.
Claude isn't the story — the loop is
Claude Code made the whole thing faster, obviously. But the interesting part isn't the speed. It's that the loop — describe, generate, run, look, adjust, show, listen, adjust — collapsed from "a few days across three tools" to "a couple of hours in one place". No handover. No exporting anything. No "can you make me a mock and then we'll build it".
"Design used to be: draw it, hand it over, wait, review, revise. Now it can be: build a rough version of it, watch someone use it, and let their reaction do the specifying."
The unglamorous prerequisite, again
None of this works from a blank folder. It works because the app already exists — the table component, the data layer, the routing, the tokens, the design system. I'm not building from scratch, I'm remixing. That is what makes two hours a realistic number instead of a fantasy one. If you don't have that foundation yet, ship a few things first and come back to this.
What you can learn from this
Show rough versions in code as early as possible. Even 40% is enough if the person in front of you is the actual user. Their reaction to something imperfect is worth more than hours of polish, because it tells you whether you are polishing the right thing.
The PR isn't the finish line. It's the moment the conversation becomes cheap to have. Everything before it — the mock, the meeting, the doc — was us trying to describe a table to each other. Everything after it is us pointing at one.
Disagree? That's the point. Tell me why.
← Back to all thoughts