How To Design In The Agent Era

Episode Summary

Executive Summary: The episode spotlights Paper, an AI-native design tool built around HTML/CSS and agent workflows, and argues that design remains a key differentiator even as AI speeds up production. Founder Stephen Haney demonstrates how Paper helps humans and agents collaborate on brand and product design, reduce handoff friction, and quickly iterate from live sites or code bases while preserving human taste and control.

Main Topics: Paper’s AI-native design architecture (Priority: 5/5): Stephen explains that Paper uses HTML and CSS as its rendering engine, making it easier, faster, and more accurate for agents to operate than tools with custom renderers. Human taste as a competitive advantage (Priority: 5/5): The conversation repeatedly emphasizes that AI can accelerate production, but human judgment, aesthetics, and intentionality are what make products stand out. Brand design, shaders, and visual experimentation (Priority: 4/5): Paper’s shader library, image generation, and font work are positioned as creative and marketing tools that let designers make distinctive, shareable brand assets quickly. Design-to-code and code-as-source-of-truth workflows (Priority: 5/5): Paper lets users copy designs as React or Tailwind, or have agents rebuild selections directly into codebases, collapsing traditional handoff boundaries. Live site import, feedback, and iterative redesign (Priority: 4/5): The product can snapshot live web pages into editable layers, then use comments and agents to generate variations, critique layouts, and refine real sites. AI design tells and how to avoid them (Priority: 5/5): Stephen identifies common model-generated design problems—over-bold typography, too many font sizes, excess cards, gradients, icons, and generic vibes—and shows how to correct them. Company building, values, and growth strategy (Priority: 4/5): He describes Paper’s growth through authenticity, community alignment, user conversations, and careful hiring, rather than feature breadth or hype.

Key Arguments: HTML and CSS are ideal for agent-native tools because models already understand them, lowering token use and improving speed and accuracy. Design is still a major differentiator; if companies rely on generic AI-generated visuals, they will blend into the crowd instead of standing out. The best workflow is not pure prompting but a hybrid of direct manipulation and agent assistance, so humans can stay in control while moving faster. A single source of truth in code reduces drift between design and implementation, eliminating a historically broken handoff process. Models are improving tactically, but taste, prioritization, and organizational decision-making remain fundamentally human responsibilities. Paper’s most valuable contribution is not just generation, but fast editing, iteration, and feedback loops that let designers curate from many agent-produced options. Paper’s internal and external growth comes from being values-aligned with designers and building genuinely useful creative tools rather than marketing gimmicks.

Data Points: Team size: about 12 - Stephen says Paper is built by roughly a dozen designers and engineers. Paper Mono launch timing: in a couple of weeks - He says the new mono font will launch soon. Old company Twitter followers: about 25,000 - He cites this as an early community-building milestone before Paper had a product. AI model version mentioned: 5.5 - Used for a live variation-generation demo and described as a balanced model. Cursor model version mentioned: Composer 2.5 - Stephen says he uses it often for speed and precision. Tooling trend timing: last December of 2025 - Stephen refers to a Cloud Code explosion that pushed Paper to build a desktop app and MCP support. Paper Snapshot: live-site content imported as editable layers - Described as replacing screenshots and enabling live editing of website sections. Model usage in image generation: four different models at once - Paper’s image generation library can run multiple models together in a variety pack. Workshop/hiring ratio: 120 FPS - Stephen says design tools need extremely fast, smooth performance for designers.

Pivotal Quotes: "Design is this great differentiator." — Stephen Haney: He argues that relying on generic AI output makes companies blend in rather than stand out. "The code is the source of truth, and both systems know how to work on that same code base." — Stephen Haney: He explains why Paper collapses the design-to-development handoff into a shared workflow. "We’re trying to create a tool that lets humans work really fast with agents." — Stephen Haney: He summarizes Paper’s core philosophy about human-AI collaboration.

Implications: Paper reflects a broader shift toward agent-assisted creation where speed matters, but taste and control matter more. For teams, the future is fewer handoffs, tighter code/design sync, and faster iteration with humans curating the final experience.