The 9 steps behind AI-enabled design​
​
Hey,

Everyone's putting AI-enabled on their process these days.

Almost nobody can tell you what that actually means.

So let me tell you what it means for me. Nine steps.

Step 1: Get the baseline. Before you touch anything, wire an MCP into your analytics. You need to know the number before you can prove you moved it.

Step 2: Audit what already exists. Connect GitHub, point Claude at the codebase, and understand what's actually built on the page you're redesigning. Not the live screen. The real thing, underneath.

Pull your Loom and session recording transcripts here too. Real users narrating real confusion is data. Feed it in and let it enrich the audit before you draw a single box.

Step 3: Map it old school. FigJam. User flows. Every edge case, drawn out by hand. AI doesn't skip this step for you. It never will. And if your process skips it, that's where your prototype quietly falls apart later.

Those same session transcripts belong here too. Half your edge cases are sitting in a Loom recording of someone getting stuck. Mine the transcripts, then map what you find.

Step 4: Go wide. Double diamond. Multiple directions, multiple prototypes. Divergence before convergence, same as it's always been.

Step 5: Validate with users. Not your team. Not your buddy. Users. Record the sessions. The transcripts become your next round of enrichment, feeding straight back into the prototype.

Step 6: Translate the winner into Figma. Once a direction's validated, use Figma's agent to screenshot your prototype and rebuild it leveraging your design system. You're not pixel pushing a single screen from scratch. You're assembling from components that already exist.

Step 7: Run design, product, and eng review together. This is where the prototype earns its keep. States you didn't think of. Statuses you can't fake in a click-through. Edge cases that only show up once something's tangible, some of them straight out of a session transcript nobody read closely until now. That's why Figma is still going to be a thing no matter how good the AI gets. Someone still has to see the whole system in one place.

Step 8: Hand off to dev.

Step 9: Ship. Then measure.

If you're not measuring, you didn't ship. You deployed.

That's the difference between a designer who says I redesigned checkout and one who says I redesigned checkout and drove $180K in recovered revenue.

One's a task. The other's a case study.

If you want the full breakdown of this process, tools, prompts, the whole workflow, that's what I teach in ROI of Design.

​Grab a spot here.

Next email? The one number every designer should be able to tie their work to.

Talk soon, Tyler

PS. None of this matters if your portfolio loses the hiring manager in two seconds flat. Nick and I roasted real, anonymized hero sections on the latest episode of The Design Table Podcast. Why lines like I deliver thoughtful designs, seamless solutions, and creative experiences built for impact tell a recruiter nothing. Why curious about X reads weaker than expert in X. And the two line hero I said needed zero notes. Watch it here:

​

Tyler White Helping designers prove the ROI of their decisions