How to Design in the Agent Era

Chapter 1 · Narration: WOW Mentor

How to Design in the Agent Era

What WOW Mentor breaks down here is a single question: when agents do the actual work, what keeps a designer indispensable? Paper founder Steven Haney built a design tool on HTML and CSS so agents can read and write it directly, and on top of that he erased the very concept of handoff. But his real message is not the technology. Do not ship whatever the model spits out; strip away the AI tells — the bold, the too-many sizes, the purple gradients; and understand that taste and judgment remain a human job. That is exactly where WOW Mentor trains the lens.

Today, WOW Mentor breaks down how to design in the agent era. The figure on stage is Steven Haney, founder of Paper, an AI-native design tool. He opens by granting the obvious: agents can speed you up, save you a great deal of work, and take on chores like translation or resizing. Then he pivots, insisting that he and his team are firm believers in the human element of design. What they are building is not a tool that pushes people aside but one that lets humans work really fast alongside agents. What WOW Mentor highlights in that first move is its order: a man who sells a tool begins not with what the tool can do but with why the human still has to be there. Conceding the limits before the powers compresses the whole story of how he won his market. He describes Paper simply. It is a design tool, but an agent-native one, and the substance of that phrase lives in the rendering engine. Where most design tools run on a custom engine, Paper uses HTML and CSS as its actual engine, because agents understand HTML and CSS extremely well and that language already sits inside the models' training data — whereas a custom engine is far harder for an agent to use, producing higher token spend, slower runs, and lower accuracy. What WOW Mentor underlines is that this is not a technical brag. He made a single strategic decision — draw the picture in a language agents already speak — and that one choice moved three levers at once: speed, cost, and accuracy. The twist is that agents were never the original point. Paper, he says, was born from love. He started as a solo founder, and it was simply the side project that kept going; there were months he wondered whether anyone would care, and then people genuinely did. The phrase he returns to is that it was the thing he was going to do anyway — a startup should begin from the strongest founder-market fit, the work you would wake up wanting to do regardless of the market. Looking around, he found you almost cannot talk about design without saying Figma, thriving in enterprise; but as Figma climbed upmarket, he asked whether it was time for a tool focused squarely on the designer. What WOW Mentor flags is an old truth: the space a giant vacates as it moves up is exactly where a new tool finds its opening. The real reason he chose HTML and CSS surfaces here too: he picked it to fix designer-to-developer handoff, letting a designer work in a constraint-free canvas while, underneath, the file is simply HTML. Then came the discovery that matters most. Build a great handoff for humans, he found, and you have also built a great handoff for agents, because ask an agent to work on a Paper file and it is literally reading the CSS out and knowing exactly what to do. That is why he calls Paper a visual interface that extends the agent — prompting is not always the best input, and sometimes dragging and drawing is simply faster, which is why sales folks and engineers reach for it. What WOW Mentor highlights is that he refused the narrow definition of a design tool and redefined the product as a visual way to talk to agents.

The passage on shaders and fonts reveals the company's character most clearly. Because Paper renders through the browser, it draws shaders — WebGL animations — with ease, and Paper Shaders, an open-source library, lets designers combine halftone dots, neuro-noise, and blending modes. What he calls this library is decisive: not product development, but the marketing budget. Instead of buying ads, build something useful for the world — it works far better than an ad. The forthcoming Paper Mono font follows the same logic. What WOW Mentor reads here is that he refuses to separate marketing from product; throwing gifts into the world is itself the growth engine. The handoff story leads into a new working stack. But the moment he names as decisive is December 2025, the Claude Code explosion. He never imagined designers would be in terminals, yet they were, so the team scrambled onto the desktop as designers began keeping local repos and learning GitHub. The new stack has several agents like Cursor and Claude Code, Paper as the visual channel for talking to them, and GitHub or code review behind that. Paper embedded an MCP server in its desktop app; the agent makes tool calls, then writes HTML directly — Paper renders it with no translation between them, and it runs both ways, rendering a real codebase's colors.css onto the canvas as a style guide or building diagrams to audit a codebase. What WOW Mentor highlights is that this single fact — never switching formats — was the core of the moment the product took off, because it is exactly what saves tokens, adds speed, and cuts hallucinations. Here he names the root disease of old handoff. A designer hands something off, and either it is not built as intended, or the code drifts out of sync — so what is the source of truth? Paper's answer is unambiguous: the code is the source of truth, and both the design tool and the code work on the same codebase. He points to the many teams that maintain two copies of a design system; keeping them in sync is not a tractable problem, which is why the more agentic newer companies treat the codebase as the single source. Paper has no components yet — a common complaint — but he makes a paradoxical case: because the codebase is the source of truth, you summon what you need when you need it. The Paper Snapshot extension pulls real sections of a live site onto the canvas as editable layers, and returning polished work is just as simple: tell the agent to build the selection into a new component and it matches your existing conventions. The handoff step disappears, leaving one loop where a single person changes, iterates, and ships. What WOW Mentor dissects here is that the answer to which side is the truth quietly determines the structure of the entire tool. The most practical passage is how to strip the AI look from a model's output. First, models love bold; pull the weights back as light as you can, and it magically looks more designed — dropping a heavy headline by a single weight instantly calms the screen. Second, models love five, six, seven, eight font sizes; work with three. Third, models often botch contrast; check whether the text is readable, and tone down supporting elements that shout too loudly. And the principle he repeats is that a large part of design is deleting. Models behave like an insecure designer, scattering meaningless little ones and twos to anchor the edges of cards — but ask whether those numbers communicate anything and the answer is no, so delete them. In Paper, deleting an element snaps the rest into place; rather than describing a headline and re-prompting, you select it and hit delete — far more control. What WOW Mentor pins down is that good design comes not from a talent for adding but from the judgment to subtract, and a tool that lets you execute that judgment fast dissolves the old trade-off between speed and quality.

… The rest of this chapter continues in the full edition, available with a subscription.

Listen to this chapter