Design Review

Stephen Haney on the handoff that serves humans and agents

Stephen Haney· Founder at Paper
·~56 min·English·Y Combinator
AgentsAI CompanyMultimodal
TL;DR

Paper founder Stephen Haney explains why building a design tool on web-native HTML and CSS turns one artifact into a shared handoff for humans and coding agents alike — and why taste and design judgment stay a human job.

01Core Mental Model

One handoff, two fluent readers

Paper renders with HTML and CSS, so the same artifact a human edits is one a coding agent can read and rewrite without translation.

if you build a great handoff for humans, you've also built a great handoff for agents, because agents understand HTML and CSS, and it's in their training data.

— Stephen Haney, Design Review
Key Insight
Paper chose HTML for a human problem — cleaner designer-to-developer handoff — and only discovered afterward that the format agents were already trained on was the same one. The moat isn't an agent feature bolted on; it's the substrate choice.
02The Engineering Payoff

Skip the translation layer

A design tool with a custom rendering engine forces agents to translate their output; a tool that renders raw HTML lets the agent write straight to the canvas.

The agent writes HTML, Paper can render HTML, and you don't have to like switch back and forth between formats. What that does is it saves you tokens. It makes it run faster, and you just get less weird hallucinations.

— Stephen Haney, Design Review
Key Insight
Every format conversion is a place for an agent to burn tokens and invent mistakes. By making the render format identical to the agent's output format, Paper removes an entire failure surface instead of trying to patch it with better prompts.
03Where Truth Lives

The code base becomes the source of truth

Teams used to keep a design file and a code base in sync by hand; agentic builders treat the code as canonical and let both designer and agent work on it directly.

this is the whole loop now. Like, there isn't even a handoff anymore.

— Stephen Haney, Design Review
Key Insight
The two-copy design system was never a tooling gap to be filled — it was a structural bug. Once the code is the single artifact, 'handoff' stops being a step to optimize and simply disappears, and with it the sync cost that made design systems expensive.
04The Right Input

A visual interface for agents

Prompting is one way to steer an agent, but for spatial design work, dragging and selecting on a canvas is often the faster instruction.

prompting isn't always the best input mechanism for agents. Sometimes you want to drag and draw and it's just faster to be able to do direct manipulation.

— Stephen Haney, Design Review
Key Insight
Most agent products assume the chat box is the interface. Haney's framing inverts that: the canvas is a richer controller for an agent than a text field, because pointing at a thing carries intent that a paragraph of prose has to reconstruct.
05Fighting the Slop

Spot the tells, then subtract

Model-generated designs share a signature — heavy weights, too many font sizes, purple gradients, cards everywhere — and most of the fix is pulling things back rather than adding.

I think a lot of design is deleting.

— Stephen Haney, Design Review
Key Insight
Paper's 'secret sauce' is just senior designers' typography and contrast rules written into the model's instructions. The tells cluster because models over-add to fill space; the reliable correction — fewer sizes, lighter weights, delete the ornaments — is subtraction, which is exactly what agents left alone won't do.
06Why Craft Matters

Design is the differentiator

As it gets trivial to ship software, most products converge on the same generated look, and the care put into design is what makes one stand out.

design is this great differentiator. And if you look at every great company of the last 10 20 years, basically all of them have exceptional design.

— Stephen Haney, Design Review
Key Insight
When the marginal cost of an app trends to zero, scarcity moves to taste. The vibe-coded default isn't just ugly — it's a signal of how little attention was spent, and users read that signal as a proxy for the care behind everything else.
07The Human Line

Models get the tactics, humans keep the judgment

Agents will keep getting better at the tactical layer of design, but deciding what is good and worth building stays a human, organizational task.

do I ever think it's going to replace a human making these design decisions? No, no way.

— Stephen Haney, Design Review
Key Insight
Haney splits design into tactics and judgment. He fully concedes the tactics to models — and even notes if every model gained taste, everything would look the same again, which only pushes the human's job up a level: finding the next thing that stands out.
08How It Was Built

The most authentic product wins

Paper reached tens of thousands of followers before it shipped a product, by being genuinely values-aligned with the designers it wanted to serve.

we didn't have a product for the first year and we still got to about 25,000 Twitter followers just by talking about our values.

— Stephen Haney, Design Review
Key Insight
Haney runs the company bottleneck-first: ask what it needs next and do the most straightforward thing — a font or a shader library as marketing, a designer talked to every day, a 12-person team that reads every line of code. In a taste-driven market, authenticity compounds where feature-count doesn't.