
Meet your instructor — a designer since 2011 — and trace how the market moved from the Photoshop era to mobile-first, Figma, and now AI. You'll see why keeping your workflow updated is no longer optional, and why this course was built.
The most important idea in the whole course: AI is the engine, you are the director. Learn why 80% of an interface's quality is decided before the first pixel, how to set "gates" between stages, and the three things AI won't do for you. Then meet the toolkit — Claude Cowork, Claude Design, three Figma plugins, Figma MCP — and the account plan you actually need.
The full roadmap: 20 steps across 8 phases, grouped into three work zones — Think (phases 1-3, all in Claude Cowork), Produce (phase 4, Claude Design → HTML) and Finish (phases 5-8, design system, Figma, MCP, prototype, handoff). By the end you'll know exactly where every lesson fits and what you'll walk away with.
A guided tour of the sample project folder — your toolkit: the client brief, assets (logo, mood board, SF Pro fonts), the three W3C token files, the three Figma plugins and the 20-step process flow. Know which folder serves which stage before the real work starts. Download the complete Pika starter kit (brief, assets, tokens, plugins, master spec) from this lecture's resources and follow along on your own machine.
Live in Claude Cowork with the real Pika brief (a pickleball court booking app): let Claude split the brief into what we know, what's missing and what's vague, confirm the assumptions table (AS1–AS7), answer the client-only questions and lock device, frame size and orientation. Rule number one: you read a brief to find the gaps, not to nod along.
Claude digs into Pika's business model, operation flow and shared terminology, scans competitors like Playtomic and CourtReserve, proposes a North Star metric — court hours booked and completed per week — and builds four personas (Min, Lan, Tom, Koa). Everything lands as concrete design implications for the home screen, availability grid, membership and rewards.
Research becomes product structure: the MVP feature list, a 5-tab sitemap (Home · Book · Bookings · Rewards · Profile) exported as a clean tree PDF, the main user flows, 60-minute slots, the 12-hour cancellation policy and the court-credit wallet. The phase closes by handing Claude the mood board to set the visual direction for the UI.
Lock the visual direction before generating anything: navy #0E203A and lime #CAF427 pulled from the logo, a bright sporty style built from tennis, fitness and workout app references, SF Pro for iOS — and a three-way asset strategy separating static, runtime and functional assets AI should never fake (real QR codes, real maps).
Consolidate everything you've built — brief, assumptions, business model, personas, sitemap, flows, content direction and mood board — into one master-spec.md, plus a ready-to-run prompt for Claude Design. One clear source of truth before a single screen is generated.
Open Claude Design, attach just two files — the master spec and the prompt — confirm the details it asks about (like the logo), then submit and watch it generate the first complete hi-fi UI for the whole Pika app.
The review loop: comment on each screen, then switch to the sharper method — screenshot the exact area that's wrong and feed it back to Claude until it's fixed. Finish by exporting the entire polished UI as a single HTML file, ready for the design-system phase.
Start a fresh Cowork task with a context summary so nothing is lost, then have Claude read the exported HTML and sync colors, typography, spacing, radius and shadows into the three W3C token files and three plugins (Import Tokens · Foundation Styles · Check) — following the rules already defined in those files, no random changes.
In Figma: import the three token files into variables (primitive → semantic → component — never skip levels), generate text and effect styles with the Foundation plugin, bring the HTML UI in with html.to.design, split every screen into its own frame by flow, then run Check → Scan Selection → Fix All to match the UI to tokens — cutting roughly 70-80% of the manual work.
Paste your Figma frame link into Cowork and let Claude, through Figma MCP, analyze the UI and extract a component list, build the library with states and variants (test three groups first), merge the button set into one component, swap instances into the screens, and fix broken spacing by adding the missing token.
The finish line: Claude wires the entire clickable prototype in Figma via MCP, then reviews it in Present mode — buttons, tabs, bottom sheets, back actions, dead-end screens — fixes what's broken and reports back. You click through every flow like a real user and package the file for developer handoff. From a rough brief to a working prototype: done. The finished Pika Figma file is attached in this lecture's resources — open it next to your own build and compare every detail.
This course contains the use of artificial intelligence.
What if you could take a rough client brief and ship a complete, hi-fi Figma prototype — on your own, in a fraction of the usual time?
That's exactly what you'll do here. No dry theory, no toy exercises. You'll watch one real project — Pika, a mobile booking app for a pickleball court chain — travel the entire journey from a vague client email to a polished, token-based Figma file with a working clickable prototype. And you'll build it with me, step by step.
I've been a designer since 2011. I've watched the market shift from Photoshop to Figma — and now to AI. Here's the truth: designers who master an AI workflow now ship in days what used to take weeks. This course hands you that exact workflow.
THE METHOD: 20 STEPS · 8 PHASES · 3 WORK ZONES
Zone 1 — Think (Phases 1-3). 80% of an interface's quality is decided before the first pixel. Inside Claude Cowork you'll dissect the brief, interrogate the business model, analyze competitors, define a North Star metric, build four personas, and turn it all into a feature list, sitemap and user flows.
Zone 2 — Produce (Phase 4). Pack every decision into one master-spec file, feed it to Claude Design with a truly visual, prescriptive prompt, refine the result with screenshot feedback, and export the entire app as HTML.
Zone 3 — Finish (Phases 5-8). The part most courses skip. Extract your UI into a real design system — three W3C token files, Figma variables, text and effect styles — then let Claude drive Figma MCP to build a component library with states, swap instances into your screens, wire the clickable prototype, and package everything for developer handoff.
Along the way you'll internalize the mindset that makes AI work for you instead of against you: AI is the engine, you are the director. You set gates between phases, you judge what's good enough, and you handle the edge cases AI won't see on its own.
By the end, you won't have a folder of scattered exercises. You'll have one complete, portfolio-ready project — and a repeatable system you can run on your very next client brief.
Note: you'll need the Claude desktop app on a paid plan (Pro at minimum) and a Figma account. No coding required.
Let's get started.