Design in the age of AI

The team didn’t use Figma. The platform changed so fast that any design file fell behind what was actually live, and a designer working there becomes the bottleneck. AI was an essential part of how engineering worked.

I kept the whole of the designer’s job: framing the problem, setting the direction, making the call on every screen and holding the bar on what shipped. Claude Code turned those decisions into the real product, which let me design at the speed the team was building.

It depended on the work and on the deadline.

For a new concept, I started on paper. I sketched in my notebook, an AI agent turned the sketch into a wireframe, and I used it to test the hypothesis before building anything. Then I built it in the real product.

For a change needed that same week, I went almost straight to building. Claude Code built the change on a branch of the real product, with the real components, and I reviewed it as a before and after with real screens.

Either way, engineering took it from there, the backend, the tests and the production details, sometimes from a ready branch, sometimes from issues with screenshots and acceptance criteria. We tried both and kept adjusting to whatever got the product updated best.

• Nothing visual ships without my eyes on it. Every change came to me as a before and after first.
• Real components or nothing. A prototype made of the real pieces shows what the product can actually do, and every workaround got a note for engineering.
• Never invent data. Screens use the real shape of the data. If the data couldn’t support a design, that became a ticket, not a fake number.
• A finding becomes a ticket. With acceptance criteria, backend included, so nothing depends on someone remembering a meeting.

I joined some customer calls, but I couldn’t be in all of them. The sales and customer success teams talked to customers every week, with people from many different companies, and most of those calls were transcribed. So we put all their notes in one shared folder, and I built agents that read them, grouped what customers said by theme and type of pain, and linked each pain to the issues already working on it.

Every week I ran it and published the result in GitHub, where the whole team could see what customers were asking for, grouped by the meeting it came from. AI became my eyes and ears in the conversations I couldn’t join, and design decisions started from what customers actually said.

When AI is part of how everyone works, the leverage isn’t drawing faster. It’s putting design judgment where the AI reads. So I wrote mine down, in my own words, and turned it into tools the team used every day.

The core of it fits on a card. These are the rules the agents apply, and the ones I apply myself.

I wrote about this way of working on LinkedIn, in English and Portuguese: what changed when a product designer stopped using Figma. It passed 1,000 reactions across both versions, and it’s still the thing people bring up first when they talk to me.

84 pull requests from a designer, 65 of them merged, and 151 issues that turned design decisions into work engineering could pick up. Three design agents the whole team used, one of them part of DS Police. A weekly loop that turned customer calls into pains tied to the work. And a way of working that let one designer keep pace with a team that worked with AI every day.