For a while, I believed the future of design would happen in code.

Vibe coding took off: describe what you want, let an agent build it, and get a working app. Tools then added ways to visually edit what had already been generated. You could adjust the layout, refine the UI, and change styles without opening a traditional design tool.

It felt like the direction everything was heading.

But something was still missing.

These tools were built to produce software, not explore design directions.

Design starts without a clear answer. You might place five ideas side by side, discard three, and combine the remaining two.

A code editor naturally pushes toward one working result: prompt, generate, refine. You can request five versions, but comparing and combining them side by side is harder.

The Canvas Gives You Room to Think

Now agents are moving onto the design canvas. You can describe a screen, component, or variation, and an agent creates it directly on the canvas as an editable visual artifact.

That brings AI into the space where designers think and explore.

Figma initially introduced a workflow for bringing interfaces from Claude Code into Figma. You can prototype an interface in code, capture the working UI as editable Figma layers, make changes on the canvas, and send those decisions back to code.

That round trip is smart. But this particular workflow still begins in code. If you’re exploring five directions before committing to one, it doesn’t quite support that part of the process.

Not long after, Paper, a design tool built on the idea that the canvas itself is code, let agents work directly on its canvas. When you prompt an agent, the result doesn’t appear as code in an editor. It appears as an editable visual artifact on the canvas.

Paper’s broader hypothesis is that designers need a visual space to think.

A chat box is good at producing one answer after another. It isn’t good at keeping five directions in view at once.

Prompt one direction, and the agent places it on the canvas. Prompt another, and it appears beside the first. Zoom out. Compare. Combine parts. Sit with the possibilities before you commit.

A canvas gives you space to think spatially, something a chat log can’t do.

Paper’s Canvas

Paper quickly drew attention for bringing agents onto the canvas. Soon after, Figma made a similar move.

Figma opened its canvas directly to agents. Claude Code, Codex, and other MCP clients can create and modify designs inside Figma files using existing components, variables, and design-system rules.

The workflow no longer has to begin in code. An agent can generate an editable UI artifact directly in the design file.

This marks a new era for design.

When Prompting Becomes Faster

What does working with an agent on the canvas look like in practice?

Let’s start with something simple: a button.

Instead of reaching for different tools and adjusting each property yourself, describe the result you want.

“Make a rounded button, 44px tall, using the primary color from my design system.”

Done.

Compare that with doing it yourself: draw a rectangle, adjust the corner radius, open the color picker, and find the right token. That’s a lot of steps for one button.

That said, if we’re only talking about speed, you might still be faster making that button manually. The agent can take a moment, especially at the beginning, while it’s loading files, understanding the context, marinating, schlepping, concocting, and doing whatever else agents do behind the scenes.

But try something more complex.

Make an order-history screen with a timeline showing each order’s status. Every step needs a timestamp, a status icon, and a connecting line. Each order should also be collapsible.

Build that manually, and you’re dealing with layout, spacing, states, icons, hierarchy, and interaction. It could take hours to build and refine.

An agent can give you a first version in seconds. It might not be perfect. It could be generic. It could be slop. All of that can be true.

But if we’re comparing the same interface, prompting can be dramatically faster than creating everything by moving a cursor.

Looking at speed alone, it seems like a no-brainer. Why would we still do things manually?

But speed isn’t the whole story. If the result is the same, does the path matter?

Draw a rectangle or type a prompt to make one. The same rectangle.

What changes when you shape something directly instead of describing what you want?

The Tool Changes How You Think

With a mouse, your hand is directly in the work. You drag, place, resize, and adjust.

With prompting, you describe what you’re thinking. AI interprets it and makes something for you.

Both are ways of creating, but they feel different.

With prompting, the loop is: think, describe, see the result.

With a mouse, the loop is: move, look, adjust.

You place a button, look at it, and think: What if it were a ghost button instead? What if I moved it to the left? What if the label were shorter?

You aren’t always imagining the finished interface first and then executing it. Sometimes new ideas emerge as you move things around. Your hands become part of the thinking process.

That makes manual work slower at times, but also more immediate. You can try something without having to explain it first.

And then there’s pen and paper. Actual pen and paper. No screen. No pixels.

I still sketch sometimes. Not because anyone is asking for hand-drawn wireframes. I do it when my thinking feels stuck. I pick up a pen and draw something rough.

I actually sketch more now when I prototype with Claude Code, often right before I write the prompt. The sketch helps me think an idea clearly before I have to put it into words.

How Close Do You Need to Be?

The way I see it, prompting, the mouse, and the pen create different distances between you and the thing you’re making. It’s a spectrum.

With prompting, you describe the outcome and let the agent fill in many of the details.

With a mouse, you make those decisions more directly. You choose where something goes, how it looks, and how it responds to everything around it.

With a pen, you’re closer to the raw idea, even if you’re farther from the finished interface. You don’t need to explain it or make it precise. No components. No design system. No polished result. Just a line on paper.

More distance can give you speed. Less distance can give you control. And sometimes getting closer to the work is how you find the breakthrough.

But the spectrum isn’t a rule. Sometimes it’s simply faster to fix something manually. Nudge this four pixels. Swap that component. Shorten the label. Done. There’s no need to write a prompt for something you can change immediately.

Some designers rely almost entirely on prompting. Some design everything manually in Figma. Some prompt a first version and refine it with a mouse. Some sketch first and feed the sketch into a prompt. Others move between all three.

There is no universal workflow. Just the one that works best for you.

So, do designers still need a mouse?

Maybe the better question is: how are you working these days?

Until next time,
Thomas

Originally posted on my Substack.