Claudemaxxing, without losing my taste
How I stopped tool-hopping, kept the thinking, and built a workflow that still looks like me.
I know. Everyone’s fed up with yet another designer’s AI workflow. Bear with me.
In my last piece, I wrote about shaping the work before you design it. This is the other half: once I knew what to build, how was I building it?
I was handing over more than tasks.
My process was all over the place. Figma for the design, Claude for the build, one tool for motion, another for video, another for sound.
And there was a sneakier problem. I’d started letting AI think for me.
If you watched the Specs launch at Config, you saw it happen live. Johnny asked Claude for a spec, reviewed it, and left comments for himself and his teammates. When it was time to make the change, he just asked Claude to do it.
Aha. That’s how sneaky it is: small decisions first, then big ones.
So I went looking.
I read every workflow I could find and talked to people. The one that clicked came from my friend Kabir, who just graduated from CIID. He never leaves Figma: he tags Claude in a Figma comment, and it makes the change and replies in the thread. Read his version →
I tried to adapt it. I wanted to stay in Figma too. But Claude Code turned out to be my home base, not Figma.
So I run on two tools. Figma for pixel control. Claude Code for everything that moves. Voiceovers come through the ElevenLabs MCP, videos from Remotion, and prototypes go straight to code.
This portfolio’s teaser, every video snippet and hover state, the type, the spacing, the design system: all built in those two places, with the timing exactly how I want it.
Fewer tools meant fewer handoffs, and more of my attention left for the decisions that should stay mine.
If the output looks like me, I’ve won.
Three things worth stealing.
Don’t worry, your girl got you. (Only to some extent; I’m learning too.)
Think before you prompt. I give myself 60 minutes with pen and paper first: users, flows, systems. Then Claude turns my sketches into clean diagrams in Figma through the MCP. The thinking stays mine.
Ask for fifteen, not one. Layout explorations used to eat my time. Now I ask Claude for 10 to 20 versions in one lab page and compare them side by side.
Write down your taste. Ask Claude to keep a taste.md file and add to it every time you react: “I like this”, “never do that”. Mine now knows my design decisions, so the work looks like Sharpi, not Claude. That’s the skill that will matter most: automate what you should, and keep your voice.
Still not there yet.
I still jump between two tools. I still catch myself asking Claude to do a 30-minute task I’d do faster by hand. Next, I’m building my own agents and internal tools, to see how far I can push this without an engineer.
The goal is work that’s an extension of me, not of the agent.
Everyone’s curious how designers work now. Here’s mine. What’s yours?