Well, kind of. How I integrated Claude into my workflow to accelerate my work while focusing on the big design decisions.
I come from an AI-first approach at Robinhood, used in a very content and redlining driven context. My exploration using Claude to start design to deployment 0 → 1 has changed my idea of what effectively working with AI looks like.
It absorbed the work I context-switch around most: work-tracking, writing, and keeping the voice consistent.
It took on the parts that need doing but eat time, so mine went to the work better spent designing.
My whole career logged out loud in one place, then mined to fill the gaps in case studies.
Fast first directions I could push against, calling out what I wanted and what I didn’t, then taking it into Figma.
Quick high-fidelity mocks showed what worked and what was missing, so I committed to a direction instead of churning versions.
Building, launching, and the QA that spots the bugs, including the mobile complexity where layouts quietly break.
Claude was harnessed to accelerate my design process, not be a designer.
Before any of this, I took an honest look at what I had, and asked friends to tell me what they really thought.
The case studies had grown bloated, and the mobile states broke constantly. It reflected an old mentality: I had been having fun with Webflow and the control it gave me over how elements move on screen, but the purposeful, intentional decisions never carried through.
Before any of the design or code, I did something that turned out to be the real foundation: I sat down and talked through my entire career with Claude as if it were interviewing me project by project. I narrated each case study the way I would in a portfolio review, and it asked the follow-ups a good interviewer asks. What was the problem. What did I actually own. What were the numbers. Why that decision and not the obvious one. What would I do differently.
The tradeoffs, the failures, the metrics had all been scattered across Figma, case studies, and pitch decks. Capturing and organizing it with Claude Cowork into one accessible place was what made rebuilding and re-evaluating the case studies possible. I set my writing rules once so everything after sounded like me.
Employing my design friends, I requested portfolio and agency websites that they thought were cool and effective.
Feeding Claude a moodboard to create a set of design principles and inspiration to start building a design language.
Dipping into Claude as an exploration first. No plan yet, just pulling the work apart.
The interesting part of this build is not that AI was involved. It is how the work moved between three different surfaces, each good at a different thing, with me steering the handoff at every step.
Claude Design tackled initial design exploration, Figma was to refine ideas on my own terms. Sometimes I’d prototype in Figma to pass to Claude Design, sometimes I’d export Claude Design to Figma. Claude in Cowork was implementation and operations: it wrote the real HTML, CSS, and JS, wired in every asset, and ran the deploy.
Rather than hand-animating every item one by one, I approached it holistically.
I also fed Claude a few examples from Figma prototypes as well as how I implemented transitions in Webflow to provide an example. Claude Design worked out the full spec to apply across the whole website.
I handed that spec to Cowork to implement verbatim, then ported it from the home page into the case-study renderer so it replays every time you navigate between projects.
For rhythm I looked at editorial, image-forward case studies, where large media and generous spacing set the pace and content reveals on scroll.
AI did not make this frictionless. The most instructive moments were the failures, and the judgment it took to recover from them. Three stand out.
Claude is less adept at importing massive amounts of design files to execute, so feeding it in pieces works far better.
Claude doesn’t like big figma files. Attempting to provide an entire figma file of interactions was a losing game.
Create foundational MDs and hand-off docs: a conversation history and a source of truth for design principles and content. That keeps Design → Cowork seamless so nothing falls through the seams.
Claude products don’t speak to each other. Passing items back and forth requires deliberate context setting.
When passing Figma to Claude, make sure each layer, group, and autolayout is named. We might not read into it as designers, but it’s part of how Claude interprets the file.
Structuring your figma file matters. Unnamed layers and bad organizational habits throw off Claude.
The least glamorous part, the web and mobile states that break, was where handing off paid off most. I set the ground rules for what should and could happen on small screens, and Claude implemented and re-tested until they held, then spent the QA pass spotting the bugs I would have missed.
The interesting work was never the layout. It was the plumbing, and the judgment calls no tool will make for you.
The takeaway is not "AI built my site." It is that AI moved my time from production to the work that actually needs a person: direction, story, and the edge cases. I made every call, and the tools removed the busywork so my attention went where judgment matters. That is the kind of team I want to build with: designers who still design, operating at the level of decisions and outcomes, fast and hands-on, but close enough to the craft to catch where the tool got it wrong.
