A recap of Promptmates Live, featuring Jamie Lyons (The Rec Lab) and Matt Parker (Move)
If you’ve spent any time building with AI over the last year, you’ve probably noticed a pattern: the outputs work, but they look like AI made them. Clunky HTML files. Generic dashboards. Candidate outreach that’s functionally solid but visually forgettable. On a recent episode of Promptmates Live, Jamie Lyons and Matt Parker sat down to make the case for a tool that fixes exactly that problem: Claude Design.
Here’s what stood out to me personally:
The Problem: AI Slop
Jamie opened with a blunt observation: a lot of the AI-built content flooding LinkedIn and recruiting workflows works, but it doesn’t land. The moment something looks obviously AI-generated, people check out, regardless of how much effort actually went into it. Engineers and analysts have known this for years about UX and dashboard design. Jamie’s argument is the same rule now applies to anything recruiters ship including things like candidate packs, decks, LinkedIn posts etc.
His framing for Claude Design: if Claude Code turned engineers into “superheroes” and gave non-engineers a way into building, Claude Design does the same thing for design: superpowering designers while giving everyone else a legitimate way to produce polished, on-brand work.
What Actually Makes It Different
A few things separate Claude Design from just asking Claude (or ChatGPT) to “make something pretty”:
It’s driven by a real design system. Before you generate anything, you set up brand guidelines: colors, fonts, voice, logos, imagery etc. You can pull this from an existing codebase, Figma files, or Jamie’s example of a shortcut is just seven or ten screenshots of a company’s website. Claude reverse-engineers a full style guide from there. This design system is doing most of the heavy lifting; skip it, and quality drops fast.
It generates code, not pixels. Unlike image generation tools, Claude Design is writing actual code behind the scenes. That means you can make surgical edits later without regenerating the whole thing and losing what was already working.
It has real editing tools, not just a chat box. Three worth knowing:
- Comment: click or circle a specific element and leave feedback, the way you’d comment on a Google Doc.
- Edit: manually adjust text, alignment, fonts, etc., without going through AI at all (useful when you know exactly what you want to say and don’t want the back-and-forth).
- Tweak: test variations side-by-side (icons, colors, toggles) without losing progress on parts you already liked.
Templates make it repeatable. Once you land on a design you like, it becomes a reusable template, meaning future outputs (a new candidate pack, a new deck) inherit the same look automatically.
It exports anywhere. Present directly, hand off to Claude Code to productionize, export as HTML/PDF, push through MCP to Canva, PowerPoint, Gamma, or Notion, or deploy straight to GitHub and Vercel.
The Live Demo: Automated Candidate Packs
The most compelling part of the session was Jamie’s live build. With a single short prompt and a Spotify-branded design system already loaded, Claude asked a handful of clarifying questions (role, format, sections to include, audience, tone) and generated a genuinely convincing candidate microsite: one version was styled so closely to a Spotify playlist page that it drew an audible reaction from the hosts.
He then walked through a more advanced setup: a custom skill (/candidate-pack [role]) that automates the entire pipeline including:
- Finds the relevant intake call notes (via a connected note-taking tool)
- Pulls out the key details: salary, reporting line, interview stages
- Duplicates a locked template (the master template is never edited directly)
- Populates it with the extracted content
- Builds it locally for review
- Pushes to GitHub and deploys live via Vercel, returning a shareable URL
For anyone not ready to wire up GitHub and Vercel, Matt and Jason floated a simpler version: stop at a polished PDF export, or run everything locally through Claude Code without publishing anywhere.
Does It Actually Move the Needle?
The panel’s evidence here was anecdotal but consistent: candidates notice and respond positively when a company has clearly invested in how outreach looks, and both Jamie and Joe pointed to upticks in response and referral rates tied to more polished candidate materials in past roles. One point from Andrea Lungallescu’s recent LinkedIn posts came up too that the more information candidates can absorb before a call, the less that call needs to be a rundown of the basics, and the more it can be an actual conversation.
A Few Other Takeaways
- Model choice (Opus, Sonnet, Fable) matters less than you’d think for design work: the real constraint is the context window, which fills up fast and eventually forces a fresh session.
- A common workflow: rough out structure in Claude Code as HTML first, then hand it to Claude Design to apply real styling: and that styling persists in future Claude Code iterations.
- Mobile isn’t optimized by default. You have to ask, and check it on your own phone.
- Claude Design is included with any paid Claude subscription, no extra cost.
Watch the Full Session
For the full hour, including the live build, the Q&A, and Matt’s parallel example of redesigning a company careers page in a week, watch the complete Promptmates Live episode below.