Y Combinator's How To Design In The Agent Era: skim's analysis identifies 12 key moments, with 2 potential conflicts of interest flagged. Paper founder Stephen Haney joins Y Combinator General Partner Aaron Epstein to demonstrate how building a design tool on an HTML and CSS rendering engine lets AI coding agents read and write designs directly.
Category: Tech. Format: Interview. YouTube video analyzed by skim.
skim AI Analysis
Credibility assessment: Practitioner Demo & Critique. Stephen Haney backs his claims about Paper's architecture with live demonstrations of HTML/CSS rendering, MCP tool calls across Conductor and Cursor, and live edits of submitted startup sites. Growth and adoption figures are self-reported by the founder or cited casually from Ramp data without independent documentation.
Bias assessment: Product-Showcase Advocacy. The episode serves as both a practical design critique and an extended showcase for Paper. Host Aaron Epstein notes that Y Combinator uses Paper internally, creating a friendly, promotional environment with minimal pushback on Paper's positioning against incumbents.
Originality: 82% — Live Agentic Design Workflow. Rather than discussing AI design abstractly, the video demonstrates live browser-to-canvas extraction and multi-agent iteration on real startup websites, paired with concrete typographic rules for stripping out AI visual clichés.
Depth: 76% — Tactical Craft & Architecture. The conversation connects underlying rendering architecture (HTML/CSS vs custom canvas engines) to token efficiency and design-to-dev handoff, while breaking down specific visual tells of AI-generated interfaces.
Key Points (12)
1. HTML and CSS Engine Powers Agent Design
Timestamp: 00:01:04 to 00:05:09
Stephen Haney explains that Paper uses HTML and CSS as its actual rendering engine rather than a custom canvas engine. Because AI agents already have HTML and CSS in their training data, they can read and manipulate Paper files directly with lower token spend, faster execution, and higher accuracy. This architecture turns a human handoff improvement into a direct visual interface for agents.
Significance (High): Using standard web primitives instead of proprietary canvas formats removes translation overhead for LLMs, positioning browser-rendered tools as natural frontends for coding agents.
Sources in support: Stephen Haney (Founder of Paper, former founder of Modulz), Aaron Epstein (General Partner at Y Combinator and host of Design Review)
2. Open-Source Shaders and Type as Marketing
Timestamp: 00:05:11 to 00:11:32
Haney demonstrates Paper's WebGL shader library, multi-model 'variety pack' image generator, and upcoming Paper Mono typeface. Rather than spending money on traditional advertising, Paper treats free creative resources like open-source shaders modeled on physical paper and a legible monospace font as its marketing budget. Aaron Epstein notes that Y Combinator uses these shaders for interactive Startup School acceptance tickets.
Significance (Medium): Shipping free, high-craft creative assets like WebGL shaders and custom fonts earns organic adoption among brand designers more effectively than paid ads.
Sources in support: Stephen Haney (Founder of Paper, former founder of Modulz), Aaron Epstein (General Partner at Y Combinator and host of Design Review)
3. Codebase as Single Source of Truth via MCP
Timestamp: 00:11:32 to 00:16:48
Following the December 2025 surge in terminal-based coding agents, Paper built a desktop app with a built-in server so tools like Conductor and Cursor can write HTML directly onto the canvas or pull a codebase's CSS into a visual style guide. Haney and Epstein agree that keeping separate design systems in sync with production code is impossible. By letting both designers and coding agents operate on the codebase, teams establish a single source of truth.
Significance (High): Bidirectional sync between local code repositories and a visual canvas eliminates the chronic drift between static design mockups and production software.
Sources in support: Stephen Haney (Founder of Paper, former founder of Modulz), Aaron Epstein (General Partner at Y Combinator and host of Design Review)
4. Legion Health Review and Multi-Agent Loops
Timestamp: 00:16:48 to 00:24:11
Reviewing psychiatry startup Legion Health, Epstein and Haney point out that generic pill badges, awkward gaps, and weak alignment make the page feel vibe-coded, hurting user trust in a healthcare brand. Using the Paper Snapshot Chrome extension, Haney copies the live DOM into editable layers, leaves a canvas comment, and directs Conductor to generate layout variations.
Significance (Medium): Extracting live webpage sections into editable layers and running overnight agent variation loops shifts the designer's role from manual layout drafting to curating options.
Sources in support: Stephen Haney (Founder of Paper, former founder of Modulz), Aaron Epstein (General Partner at Y Combinator and host of Design Review)
5. Fixing AI Slop by Cutting Font Weights
Timestamp: 00:24:11 to 00:28:11
To strip out generic AI styling on the Legion Health variations, Haney shares two immediate rules: reduce heavy bold font weights and limit any layout to at most three font sizes instead of the five to eight sizes models typically output. He and Epstein also delete decorative step numbers that add excessive contrast without aiding comprehension. Once refined by hand, the updated design can be written straight back into the codebase as a component by the coding agent.
Significance (High): Applying strict typographic constraints—lightening font weights and capping font sizes at three—gives founders a fast, repeatable formula to make AI-generated pages look human-crafted.
Sources in support: Stephen Haney (Founder of Paper, former founder of Modulz), Aaron Epstein (General Partner at Y Combinator and host of Design Review)
6. Sytex Critique and Typography Guardrails
Timestamp: 00:28:27 to 00:32:42
Examining field-operations startup Sytex, Epstein and Haney note that its heavy bold headlines, purple glows, and buried construction video obscure what the product actually does above the fold. Haney grabs a text-heavy section with Paper Snapshot and prompts Cursor to clean up the frame using at most three font sizes and regular-or-lighter weights. He reveals that Paper's own internal prompts simply encode these senior-designer rules of typography and contrast as guardrails for the models.
Significance (Medium): Encoding basic typographic guardrails into system prompts prevents AI models from defaulting to cluttered hierarchies while keeping the core product message legible.
Sources in support: Stephen Haney (Founder of Paper, former founder of Modulz), Aaron Epstein (General Partner at Y Combinator and host of Design Review)
7. Epstein and Haney Detail Top AI Design Tells
Timestamp: 00:33:51 to 00:37:39
Haney warns founders against shipping raw AI design output because every great company of the past 20 years has used exceptional design to stand out and recruit talent. Together with Epstein, he catalogs the classic tells of vibe-coded interfaces: two-pixel colored card borders, excessive card grids, purple gradients copied from Linear, all-caps spaced kickers, pill badges with icons, and unnecessary light/dark mode toggles on MVPs.
Significance (High): Cataloging the exact visual clichés of LLM-generated UIs gives founders a concrete checklist of elements to strip out before launch so their product does not look disposable.
Sources in support: Stephen Haney (Founder of Paper, former founder of Modulz), Aaron Epstein (General Partner at Y Combinator and host of Design Review)
8. Balancing Energy and Trust on Moreta's Site
Timestamp: 00:37:39 to 00:41:11
Reviewing QR payments wallet Moreta, Epstein remarks that a financial product looking vibe-coded makes users nervous, though Haney praises its energetic international vibe and interactive corner QR code. Showing a GPT-5.5 redesign in Paper that made the site cleaner but stripped away its excitement, Haney explains that models can lay a foundation, but humans must dial back text gradients and bring key proof points like the 'How it works' section and QR demo into the hero.
Significance (Medium): Automated redesigns often over-correct by flattening a brand's personality, proving why human designers must manually balance visual energy with financial credibility.
Sources in support: Stephen Haney (Founder of Paper, former founder of Modulz), Aaron Epstein (General Partner at Y Combinator and host of Design Review)
9. Why AI Agents Cannot Replace Human Taste
Timestamp: 00:41:11 to 00:43:56
Asked whether AI agents can learn taste, Haney predicts models will steadily improve at tactical execution like letter spacing and avoiding excessive bolding, citing Fable as an encouraging step. However, he argues agents will not replace human designers because real design is an organizational decision-making process involving stakeholders, requirements, and market positioning, potentially taking 10 years for models to approach.
Significance (High): Separating tactical pixel execution from strategic product judgment explains why AI tools increase rather than eliminate the value of human taste.
Sources in support: Stephen Haney (Founder of Paper, former founder of Modulz), Aaron Epstein (General Partner at Y Combinator and host of Design Review)
10. Every Line of Core Code Is Human-Reviewed
Timestamp: 00:43:56 to 00:46:52
Haney reveals that while Paper's 12-person team uses Cursor, Claude Code, and BugBot extensively, multiple humans still read every single line of production code because building a 120 FPS design tool requires extreme precision. In contrast, for marketing and peripheral projects, brand designer Agu used Paper and coding prompts to build and ship Paper's entire interactive Next.js website in a single week by himself.
Significance (High): Distinguishing between core engine code (requiring 100% human review for 120 FPS performance) and marketing surfaces (shipped solo by a designer with agents in a week) offers a pragmatic blueprint for engineering teams.
Sources in support: Stephen Haney (Founder of Paper, former founder of Modulz)
Neutral sources: Aaron Epstein (General Partner at Y Combinator and host of Design Review)
11. Building a 25,000-Follower Cult Before Launch
Timestamp: 00:47:16 to 00:50:40
Addressing how Paper built a cult following and reached parity with Sketch on Ramp's corporate card spend data behind Figma, Haney credits strict values alignment with designers. Before having a product in its first year, Paper grew to roughly 25,000 Twitter followers simply by talking authentically about design and typography on podcasts.
Significance (Medium): Building an audience around shared craft values before writing product code creates a forgiving community of early adopters, provided the founder focuses on one clear user persona.
Sources in support: Stephen Haney (Founder of Paper, former founder of Modulz), Aaron Epstein (General Partner at Y Combinator and host of Design Review)
12. Daily User Talks and Canvas Prototype Iframes
Timestamp: 00:50:44 to 00:55:35
Sharing lessons from his second startup, Haney highlights 'bottleneck-first' thinking and his rule of talking to a designer every single day during Paper's first year, a habit now required of all 12 employees on Discord and Slack.
Significance (Medium): Embedding live Vercel and HTML prototypes inside a shared design canvas solves the fragmentation created by vibe-coding, turning scattered links into a structured review workflow.
Sources in support: Stephen Haney (Founder of Paper, former founder of Modulz), Aaron Epstein (General Partner at Y Combinator and host of Design Review)
This analysis was generated by skim (skim.plus), an AI-powered content analysis platform by Credible AI. Scores and classifications represent the platform's AI-generated assessment and should be considered alongside other sources.
