Back to Y Combinator Startup Podcast

How To Design In The Agent Era

Y Combinator Startup Podcast

Full Title

How To Design In The Agent Era

Summary

This episode features Stephen Haney, founder of Paper, an AI-native design tool.

The discussion focuses on how Paper leverages AI agents and HTML/CSS to enhance the design process, improve developer handoff, and assist in creating standout designs in the agent era.

Key Points

  • Paper is an AI-native design tool built using HTML and CSS as its rendering engine, making it highly compatible with AI agents and reducing token spend and increasing accuracy compared to tools with custom rendering engines.
  • The core philosophy behind Paper is its love for design and building creative tooling, aiming to be a great design tool for humans that also works exceptionally well with AI agents.
  • Unlike many traditional design tools, Paper's use of HTML and CSS as its rendering engine facilitates a seamless design-to-developer handoff, as agents understand this format intrinsically.
  • The tool offers features like "Paper Shaders" for creating modern, animated effects, an image generation library utilizing multiple models, and a font "Paper Mono" designed for broad application.
  • Paper addresses the bottleneck of inspiration and ideation in the creative workflow by enabling rapid exploration of design possibilities and generating variations through AI.
  • The platform emphasizes the importance of human oversight and curation in AI-generated designs, highlighting that while agents can accelerate tasks, human taste and strategic decision-making remain crucial for exceptional outcomes.
  • Paper's approach to design-to-developer handoff treats code as the source of truth, eliminating the traditional disconnect and maintenance costs associated with managing separate design and code systems.
  • The tool incorporates features like "Paper Snapshot" to capture live website content for editing and a Chrome extension to bring live sites directly into the Paper canvas for iteration.
  • The discussion critiques common "AI tells" in design, such as excessive boldness, multiple font sizes, poor contrast, and overreliance on gradients and icons, advocating for intentional design choices to make a brand stand out.
  • Paper aims to empower designers by providing tools that integrate seamlessly into agentic workflows, allowing for faster iteration, better quality, and a more direct path from concept to shipped product.
  • The founder emphasizes that while agents can accelerate processes, the human element of design, encompassing strategic decision-making, understanding organizational needs, and conveying value, remains irreplaceable.
  • Paper's growth strategy relies heavily on being values-aligned with its market, fostering authenticity, and building a community around a shared love for design, which has led to significant organic adoption.
  • The company is continuously developing features such as components and prototyping/feedback tools to address user needs and bottlenecks, aiming to balance AI acceleration with human-driven precision and care.

Conclusion

AI agents can significantly accelerate design tasks and improve workflows, but human creativity, taste, and strategic decision-making remain essential.

Tools like Paper, built with an AI-native approach and a focus on seamless human-agent collaboration, are key to navigating the future of design.

Prioritizing user needs, maintaining design authenticity, and focusing on core product value are crucial for building impactful and enduring design tools and companies.

Discussion Topics

  • How can designers effectively collaborate with AI agents to push creative boundaries while maintaining brand authenticity?
  • What are the most significant "AI tells" in current web design, and how can designers overcome them to create more trustworthy and intentional digital experiences?
  • As AI becomes more integrated into the design process, what will be the enduring role of human creativity and strategic decision-making in shaping the future of exceptional design?

Key Terms

Agent-native
Software or tools designed from the ground up to be used by or with AI agents.
HTML
Hypertext Markup Language, the standard markup language for documents designed to be displayed in a web browser.
CSS
Cascading Style Sheets, a style sheet language used for describing the presentation of a document written in HTML or XML.
Rendering engine
The component of a web browser that takes the described content and displays it on the screen.
Token spend
In AI, the cost associated with processing input and output data, often measured in tokens.
Founder market fit
The degree to which a founder's passion and skills align with the market need for their product.
Vibe-coded
A design aesthetic that is trendy or fashionable, often associated with a particular online subculture or style.
MVP (Minimum Viable Product)
The version of a new product that allows a team to collect the maximum amount of validated learning about customers with the least effort.
Fetch API
A modern interface for making asynchronous network requests in JavaScript.
WebGL
A JavaScript API for rendering interactive 2D and 3D graphics within any compatible web browser without the use of plug-ins.
Shaders
Small programs that run on the GPU to perform graphics calculations, often used for visual effects and animations.
Halftone dots
A printing technique that uses dots of varying size or density to create the illusion of continuous tone.
Neuronoise
A type of visual texture or pattern that mimics natural or digital noise.
PNG (Portable Network Graphics)
A raster-graphics network graphics file format that supports lossless data compression.
Tailwind CSS
A utility-first CSS framework for rapidly building custom user interfaces.
React component
A reusable piece of UI code in the React JavaScript library.
IDE (Integrated Development Environment)
A software application that provides comprehensive facilities to computer programmers for software development.
MCP (Model Configuration Protocol)
(Likely a proprietary term used by Paper for agent communication within their tool).
JSON (JavaScript Object Notation)
A lightweight data-interchange format.
Hallucinations (AI)
When an AI model generates inaccurate, fabricated, or nonsensical information.
Style guide
A document that contains the design standards and practices for a brand, including colors, typography, and spacing.
Artboards
In design software, separate areas within a document where different designs or layouts can be created.
Sub-agents
Smaller AI agents that can be deployed by a primary agent to perform specific tasks.
Prompting
The act of providing input or instructions to an AI model to elicit a desired output.
Above the fold
The portion of a webpage that is visible to a user without scrolling.
MVP (Minimum Viable Product)
The version of a new product that allows a team to collect the maximum amount of validated learning about customers with the least effort.
YC (Y Combinator)
A prestigious startup accelerator program.
Radix UI
An open-source component library for building React applications.
WorkOS
A company that provides enterprise identity solutions.
Bottlenecks
Constraints or impediments that slow down a process or workflow.
Discord
A communication platform popular with online communities.
Slack
A collaboration hub for communication and work.
FOMO (Fear Of Missing Out)
A feeling of anxiety that arises when one feels they might be missing out on a positive experience or opportunity.
Pedantic
Excessively concerned with minor details or rules.
Fetch API
A modern interface for making asynchronous network requests in JavaScript.

Timeline

00:01:06:920

Paper is introduced as an AI-native design tool using HTML and CSS as its rendering engine.

00:02:08:560

The founder discusses Paper's origin from a love for design and the desire to build a next-gen tool that's agent-native.

00:03:17:640

The concept of being an "AI-native, agent-native" design tool is explained through its use of HTML and CSS.

00:05:19:520

A demonstration of Paper's features begins, starting with visual elements like shaders.

00:06:36:234

The ease of creating animated effects and customizing shaders in Paper is highlighted.

00:08:27:194

The image generation library within Paper, capable of using multiple models, is showcased.

00:09:24:354

Features like vectorizing and extracting colors are demonstrated.

00:10:25:674

The concept of treating creative assets like shaders as a marketing budget is discussed.

00:10:37:994

The launch of Paper Mono, a new font, is announced as another marketing effort.

00:11:32:594

The tool's origin in solving design-to-developer handoff issues is revisited with a demo of copying code.

00:12:28:794

The emergence of cloud code and the transition to a desktop app for Paper are discussed.

00:12:57:908

Paper is positioned within a new "agent stack" as a visual interface for communicating with agents.

00:13:50:268

A live demo illustrates how Conductor can interact with Paper to create a test frame.

00:14:51:988

An example of using Cursor to bring CSS styles into Paper as a style guide is shown.

00:15:31:388

The persistent problem of design-to-developer handoff and the code-as-source-of-truth approach are discussed.

00:16:37:588

The Chrome extension allowing users to grab live sites and paste them into Paper is mentioned.

00:17:07:228

The segment reviewing user-submitted sites begins with Legion Health.

00:18:39:908

Paper Snapshot is demonstrated as a tool for grabbing live site content for editing.

00:19:21:142

An agent is prompted to generate variations of a selected design section in Paper.

00:20:42:862

The ability to leave comments within Paper for agents to address is showcased.

00:21:33:862

The iterative nature of AI design assistance and its self-correction capabilities are discussed.

00:22:58:782

The host asks about go-to models, and Paper's approach to using different models for various tasks is explained.

00:24:19:222

The discussion shifts to how humans can refine AI-generated designs and avoid common AI tells.

00:25:33:782

Simple but effective design tweaks, like adjusting font weights and sizes, are demonstrated.

00:26:33:542

The process of wrapping refined designs back into code is explained, including copying as React or asking the agent to build a component.

00:27:29:341

The overall loop of design, iteration, and shipping with Paper is highlighted.

00:28:26:581

A second user-submitted site, Cytex, is reviewed and critiqued.

00:30:27:621

The agent is tasked with cleaning up the Cytex frame by reducing font sizes and removing bolding.

00:31:26:101

The effectiveness of providing guardrails and specific instructions to models in Paper is discussed.

00:32:46:389

Common AI tells in design are identified, such as overusing bold, multiple font sizes, and gradients.

00:35:40:590

The concept of "best tool for the job" is applied to design, emphasizing that both prompting and direct manipulation are valuable.

00:37:41:431

A third user-submitted site, Moretta, is reviewed, with initial critiques and then a demonstration of Paper's improvement.

00:40:46:080

The critical role of human control and the ability to edit AI-generated designs in Paper is emphasized.

00:41:03:840

The core belief in the human element of design, even when accelerated by agents, is reiterated.

00:41:15:840

The discussion turns to whether agents can learn "taste" and the future of AI in design.

00:43:58:960

How Paper uses agents internally for various tasks, from coding to marketing content, is shared.

00:44:42:127

The importance of a small, elite team and using agents to accelerate people is discussed.

00:45:48:689

A specific example of a brand designer using Paper to build the company website is provided.

00:46:43:768

The company's commitment to quality and care, even if it means moving slower, is highlighted.

00:47:36:008

Paper's growth strategy, focusing on values alignment and community building, is detailed.

00:49:43:008

Lessons learned from previous company ventures are contrasted with the current approach at Paper.

00:50:49:888

The philosophy of identifying and addressing bottlenecks, starting with early adopters and then moving to feature development, is explained.

00:51:47:007

The practice of talking to users daily and its importance in guiding decision-making is emphasized.

00:53:10:968

The future of design in organizations is discussed, focusing on the enduring importance of the design function.

00:54:38:055

What's next for Paper, including organizing prototypes and feedback cycles, is revealed.

00:55:38:215

The episode concludes with thanks to the guest and a sign-off.

Episode Details

Podcast
Y Combinator Startup Podcast
Episode
How To Design In The Agent Era
Published
August 7, 2026