Hey Reader,

I’ve been fielding the same question in nearly every coaching call lately: now that AI can spit out a polished prototype in minutes, how do you tell good taste from a lucky guess? This issue leans into that tension, starting with how one community member used research to walk into an engineering debate with conviction instead of a hunch.

In today's email:

• Research first, design second: how Andy from Level up Club turned my AI research workflow into a skill, and the simple routine you can try on your next screen.
• Scaling taste when anyone can ship: Wealthsimple's VP of Design on "design drift," the taste.MD file she built to keep AI and humans shipping to the same standard, and a podcast chat on staying creative once AI already has the answer.
• AI tools for craft, not just speed: a motion design prompt pack for vibe-coded apps, and a reference library built around the small details (not the whole page) that make a site feel considered.
• Career and wellbeing check ins: a tool that scores job postings before you apply to them, and a new annual survey measuring how designers are actually holding up.

A lot of you have been asking for it... in my latest video I share a roundup of AI-assisted design and build workflows I’ve been experimenting with lately — what’s actually sticking, what I’m still testing, and how it’s all starting to fit together. Watch the video →

Design gems of the week

• ​Teaching AI agents taste: Wealthsimple's VP of Design Polly D'Arcy on the challenge of shipping faster: "design drift," and the TASTE.MD file she built so AI and humans ship to the same judgment.
  ​
• ​Staying creative when AI has all the answers: A conversation with Ricardo Vázquez, design leader at Stripe, on keeping play and creativity alive on a team now that AI can generate an answer for almost anything.
  ​
• ​UI Motion Prompts: A collection of ready made AI prompts and React code for the hover states, loaders, and toggles your vibe-coded app is probably missing.
  ​
• ​Details.so: A library of design inspiration organized by the actual pieces (hero sections, footers, transitions) instead of full page screenshots, so you can find a strong reference in minutes instead of hours.
  ​
• ​applyjobs.club: Scores job postings against your resume, tailors your materials for the strong matches, and tracks the results.
  ​
• ​Design Wellbeing Index: A new annual, anonymous survey on burnout, job security, and what is actually helping designers cope.

Open design roles:
​Lots this week! 👀

• ​Director of Design (Propel)- Remote, USA
• ​Product Designer, tvOS (Apple) - Cupertino, California, USA
• ​Product Designer, Apple Weather (Apple) - Culver City, California, USA
• ​Senior Product Designer, Credit & Cards (Finom) - Remote, EU
• ​Lead UX Designer (Ryan) - Hybrid, Dallas, TX or Burlington, MA
• ​Staff Product Designer (StackAdapt) - Remote, CAN or USA
• ​Designer (Ambition) - On-site, Toronto, CAN
• ​Senior Product Designer II, Growth (Splice) - Remote, USA
• ​Senior Staff Product Designer (Apollo GraphQL) - Remote, USA
• ​Senior Product Designer, New Verticals AI Products (Playlist) - USA
  ​

A MESSAGE FROM RENDER COACHING You're good at the work. Where is it taking you? ​Lead by Design is a 90-day coaching program from Nicole Pikulin, a leadership coach and former design director with 15+ years leading design teams. In six sessions you'll see how your energy shows up under pressure, write your Superpower Statement, map the role you want next, and leave with a 90-day action plan. Subscribers get 10% off with code FEMKE10 through November 15. Book a free intro call →

Research first, design second: how to stop AI from guessing

A couple of weeks ago, Andy, a product designer inside Level up Club community, sent me a Loom.

He'd watched my three-part video series on using AI to research before you design, and built the workflow into a skill. It takes a project, asks a few questions, pulls in references from real products, and writes a brief.

He recently used it on a confusing flow where engineers were debating and defending their own ideas. He explored a few directions, brought them to the team, and recommended one. In his words, it gave him the conviction to pick a direction on a project where he wasn't sure where to start.

AI can reason, but it hasn't seen what's shipped

​In the first video, I gave Claude a PRD for a made-up challenger bank and asked it to design the onboarding flow. It did a pretty good job. The sequencing made sense and it thought about the anxious moment where you hand over your ID.

What it couldn't tell me was how competitors handle different moments; how Monzo handles that moment compared to Revolut, how much copy is too much, or what the waiting screen looks like in a product that's converted well.

Without real examples to work from, AI fills those gaps with assumptions, which is a big part of why so much AI-generated UI looks the same.

The fix isn't a better prompt. It's better context.

Turn references into decisions

When I redesigned the feedback screen in my own app, I researched in two passes: what good feedback says, and how it's built (components, layout, what's shown first).

What stood out was how rare specific, actionable feedback is. Most tools either stay vague or rewrite your work for you. That gap ended up shaping the whole redesign.

​For the final episode, I designed an onboarding flow from scratch. Before touching a screen, I had Claude turn the research into a pattern brief: a short list of decisions about what to collect now versus later, where goal-setting goes, and how progress gets shown. Every decision pointed back to a real product that had shipped it.

The brief made the AI's output better, because every screen was built against the same decisions. It also makes the work easier to share. If a few designers each build part of the same flow from their own prompts, you end up with screens that are fine on their own but don't agree with each other. A shared brief keeps them consistent, whoever builds them and whenever.

Why it changes the conversation

"I think it should work like this" is easy to argue with.

"Here are five products that handle this moment, here's the pattern, and here's where I think we should do it differently" is a lot harder to dismiss. That's what Andy had going into his engineering debate.

It doesn't replace your judgement, though. A pattern that works for Revolut might be wrong for your product. Research gets you to informed inputs faster. What you do with them is still up to you.

Try this week

Pick the next screen or flow on your plate. Before you ask AI to design anything:

• Research first. Ask how 3–4 shipped products handle this exact moment, on the same platform you're designing for.
• Turn it into decisions. Ask for 3–5 concrete rules, each pointing to a specific reference.
• Check the references yourself. Look at the actual screens, not just the summary.
• Bring the brief to your next review, alongside the design.

If you want to see the whole workflow, here's the series:

• ​Episode 1: Why AI-generated UI falls short​
• ​Episode 2: Fixing a real screen in my own app​
• ​Episode 3: Building a flow from scratch with a pattern brief​

Thanks to the folks at Mobbin for sponsoring the series. I use their MCP for the research step, but the approach works with any references you can give your AI tool.

If you've built your own version of this workflow, hit reply. I'd love to see it.

Have a topic you'd like me to write about? Reply to this email and let me know! I read every reply.