Back

Margin

Timeline

5 days

Team

Luis Somasundaram

Sarah Kwakkelaar

Tools

Claude Code

Figma

Skills

Research

Product Design

AI Prototyping

Impact

We flew to NYC to present Margin to a room of designers from OpenAI and Anthropic

Margin was selected as 1 of 5 finalist teams out of 800+ participants in the SVSD Design Competition. The crit sparked a lively conversation about how designers reflect on their process, and how AI can help us see it more deeply without losing the intention behind it.

Team presenting Margin at the SVSD finals

Context

Coding agents let designers iterate faster than ever

Before AI, design lived mostly on the Figma canvas, where slow, deliberate micro decisions shaped every visual and interaction detail. Now designers work in coding agents with only some starting in Figma first.

Figma

Claude Code

Problem

This new shift makes it easy to move fast without leaving a trail

The speed of AI comes with a tradeoff: we know our designs less intimately. The faster we iterate, the faster earlier decisions fade. We have iteration 712 but can't remember the decision we made at iteration 21.

Solution

Margin: a Mac app that helps designers document decisions as they work without losing speed

Margin lets designers capture iterations and decisions in the moment, all in one place. We preserve designers’ intention by capturing their thinking and scaffolding the storytelling afterward.

Capture

Select a Figma frame, text, or a URL and double-press ⌘ to capture it. Nothing selected? Double-press ⌘ for a screenshot, recording, or Claude Code capture.

1 of 7Figma frames

Annotate

After a capture, a small popup lets you jot down thoughts or decisions, so the why stays with the artifacts instead of fading later.

Remember

Margin quietly autocaptures in the background, keeping only moments with meaningful changes so nothing valuable fades. If a while passes without a capture, a gentle nudge invites you to add one.

Archive

In this chronological view, zooming in and out groups artifacts into sequential chunks of the design process, each summarized by our intelligence layer.

Screenshots are flat. So we embedded interactive Claude Code builds, websites, Figma flows, and motion.

Themes

Our intelligence layer first sorts artifacts into general themes as a starting point. From there, designers make sense of their work by moving artifacts, creating new groups, and renaming them.

Handoff

When designers are ready, they hand their organized work to their coding agent and decide how it can help, e.g. scaffold a case study, build a slide deck, prepare a pitch.

Research

Documentation is often left to the end, leaving designers to fill in gaps from memory

We interviewed 5 Master of Design students at UC Berkeley. All of them said documenting their process matters for understanding their intention. Yet none of them actually do it during the process.

A grid of faint empty slots with only four filled in

“Honestly, I forget most of the time”

Boxes scattered at random with no order

“Such a mess. It's everywhere right now.”

A grid of boxes with dashed gaps where some are missing

“I end up having to fill in some gaps.”

We reflect while we work but lose the insights by the end

As Donald Schön coined it: reflection in action, or the thinking we do mid work. A case study is about the decisions behind the final product, but we rarely document them along the way. With AI, there’s even more to lose.

Decision 1

How do we make capturing feel as seamless as possible?

Designers often leave documentation for later because it feels like a burden. What format? Where does it go? How do I also keep my thinking?

We first looked at what designers would want to capture (Figma frames, notes, Claude Code builds) and the familiar instincts like Shift-⌘-4 for screenshots.

While this is familiar, two questions came up. Could we make it faster? Could we go beyond a screenshot to capture formats that stay useful when we reflect on them?

Hotkeys are fast, so we tried a different one for each artifact type.

But having to remember which key goes with which artifact added more cognitive load.

We landed on a universal hotkey for all artifact types: double-press ⌘. Like a camera, you select what you want and capture. If nothing is selected, it falls back to a screenshot, recording, or Claude Code capture.

Decision 2

How do designers want to visualize their documentation?

We thought about how to make this messy, nonlinear design process, well…linear. A chronological view could show how the project evolved. But evolution isn’t just more artifacts being added—it’s also how the ideas behind them progress.

The chronological view was too flat and didn't account for the depth of the process.

We explored a spatial view organized by idea. The deeper you zoom, the finer the iterations. But it got messy as artifacts piled up, and zoomed out, too much was abstracted away.

So we came back to the timeline, since seeing the process in order mattered. To show how ideas progress, designers can zoom in and out, and our intelligence layer summarizes each stretch of time.

Decision 3

What does making sense of this documentation look like?

We explored having Margin write the case study from all the documentation and annotations. But our goal was to safekeep the designers’ intention—not automate away what makes the process personal.

We asked a UC Berkeley MDes professor how she'd feel about handing sense making over to AI. She said:

“Sense making is the creative part. Why would I give that away?”

So instead of doing it for them, we designed a canvas to get messy on. Our intelligence layer gives it a first pass, and from there, the sense making is theirs.

Feedback

The crit opened up a wider discussion on how AI could help tell the story

Designers from OpenAI, Spotify, and Cosmos pushed us on how AI could help tell a targeted story from the documentation while the intention stays with the designer. For example, tracing the throughline across specific screenshots, or gathering materials for an interview focused on visual design.

Master of Design students used Margin during their projects, and many said capturing felt so seamless that it would be a disservice to themselves not to.

Designers listening to the Margin presentation at the SVSD finalsPointing out an artifact in Margin's archive on a laptopOver-the-shoulder view of a laptop during the SVSD finals

Reflection

AI is changing how we design, but we can’t lose what made us fall in love with it

At its core, being a designer means being deliberate about our decisions and keeping our users at the center. I’m grateful to have met such talented designers who inspire me to keep raising the bar. At the same time, I’m thinking about where design is headed and how to move fast while staying grounded. And Margin is our attempt at starting that conversation.

Walking through New York with the SVSD finalist teams

Next

Pseudorandom

Integrating transparent AI workflows for architects

I love what I do, I love who I do it with