A Scholarly Analysis of Agent-Based Design: Paper vs. Claude Design
This video presents a live, comparative evaluation of Paper—an AI-native design canvas controlled via Claude Code—against Claude Design, hosted by Riley Brown with design expert Katherine Labry. The investigation focuses on workflow efficiency, design quality, and tool integration for agent-driven creative tasks.
🛠️ Key Tools and Workflows Demonstrated
- Paper + MCP Server Architecture: Claude Code controls Paper through a Model Context Protocol (MCP) connection, enabling direct manipulation of design elements on the canvas
- Scraping Skills Integration: Claude Code can scrape web assets (e.g., YouTube thumbnails via VidIQ) and import them directly into Paper frames for reference and iteration
- Thumbnail Generation Workflow: The agent imports 12 top-performing thumbnails from a creator, then Paper’s internal AI generates 7 variations by replacing faces, text, and backgrounds—allowing rapid ideation
- Presentation and Deck Building: A 6-slide sponsor deck was created by combining scraped photos, brand styling (Perplexity aesthetic), and real-time data from a personal website, then deployed as a live site via Vercel
- Cross-Tool Comparison: The same three-slide deck prompt was executed in both Paper and Claude Design, measuring design output, speed, and editing flexibility
- Multi-MCP Chaining: Another workflow connected Mobin MCP (design inspiration) with Paper MCP to visualize competitor references side-by-side
📊 Live Comparison Verdict: Paper vs. Claude Design Paper produced superior design quality (scored 7–8/10) with more varied templates and polished typography, but required ~40 minutes and a separate $20/month subscription. Claude Design was significantly faster (under 2 minutes for a comparable deck) and cost-included, though its designs were less refined (e.g., unwanted "bootstrapped" text appeared). Katherine preferred Claude Design for speed and cost efficiency, while Riley valued Paper for its direct pixel-level editing capability—a feature Claude Design lacks, allowing iterative tweaks without reproaching the AI.
Final Takeaway: Neither tool fully replaces the other. Paper excels for high-fidelity, research-heavy visual projects (thumbnails, decks with scraped assets) where manual polishing is needed. Claude Design is better for rapid, low-cost prototyping when design quality is acceptable. The choice hinges on whether your workflow prioritizes speed/price or granular control over the final aesthetic.