Structured in black and white first, styled from a taste profile they built themselves, then launched — SEO and OG image handled. Their product, never a toy.
The premise
Most people in these rooms are not strong at design, and that was true before AI. What AI changed is the feeling: it makes you believe you understood everything and made progress. That is why AI-built pages come out sloppy — not only the visuals, but the content. Nothing on the page is clear because no design decision was actually made.
Design is what makes it clear. So the session sequences the decisions: structure and words first, in black and white. Taste second, collected deliberately. Application third, once both are locked. Launch last. The AI executes at every step; the attendee judges at every step.
Format
Teaching blocks are 30 minutes or less, and every block ends in a do-it-now exercise. Everyone works on their own page — a homepage, a landing page, something real of theirs. Never a sample project.
The first 30 minutes delivers the win: their page, live on a URL, as a black-and-white wireframe. Real, theirs, deployed before the first coffee refill. It also proves the premise on the spot — their page reads more clearly in black and white than their last AI attempt did in full color.
The pipeline
Slop checks recur at every stage. Structure locks before style arrives — so later choices are design decisions, not a lottery.
Build the page as a strict black-and-white wireframe. With visuals off the table, iterate on structure and write the real words — actual headline, actual value proposition. A no-slop check runs on the result: unclear content has nowhere to hide.
Done looks like A reviewed wireframe with real content they can defend section by section.
Hunt references that genuinely speak to them — sites, typefaces, color directions — and collect them with a one-line reason for each. Preview what those picks imply (type samples, color pairings, primaries) before anything touches the page. Taste stops being abstract: finding what speaks to you and saying why is the skill.
Done looks like A locked taste profile: references, reasons, type and color samples.
Hand the taste profile to their AI as a plain design file. The AI produces two or three styled options of the same wireframe. Because structure was locked first, the options differ in taste, not in content — comparing them is finally a design decision. Slop checks continue so nothing creeps back in during styling.
Done looks like 2–3 credible directions of their page, in their taste; one chosen.
Refine the chosen direction section by section — small, judged passes instead of one giant regeneration. Alongside, extract a design library for the new site: tokens, components, patterns. That library is the harness: you are not prompting harder, you are building the rails the agent runs on so it stops re-inventing styles on every edit.
Done looks like A refined page plus a small design library they understand and own.
Run the launch pass for the technical tail: SEO basics, the launch checklist, and the OG image. Then publish — the AI deploys to whatever they already use, Cloudflare, Vercel, or anything else.
Done looks like Their page, live, launch-checklist clean.
By the end of the session, every participant has:
Model-agnostic
Every prompt and handoff file is plain text. Attendees use whatever they already work with — Claude, ChatGPT, Cursor, Gemini, anything. Nothing in the session assumes one vendor.
For reviewers
Internal review with Madhuri, Rajat, and Arjun — not a public checkout. Feedback on the premise, the pipeline, and the first-30-minute win.
FAQ
One hands-on session. Teaching blocks stay ≤30 minutes; every block ends in a do-it-now exercise. Not a tool tour — a sequenced thought process with a live black-and-white wireframe in the first 30 minutes.
Professionals who already want to use AI or are actively exploring it — slightly informed, not raw novices. They walk out having practiced the judgment steps they were missing: structure, taste, application, launch.
Their tagline is “Build with AI. Judge with taste.” This session teaches the judgment half on a real page — complementary to the “build with AI” half, not a replacement.
No. Prompts and handoffs are plain text and model-agnostic. Use Claude, ChatGPT, Cursor, Gemini, or whatever you already use.