Field notes

Stop Drawing Paths. Start Building Loops.

The designers pulling ahead aren’t prompting harder. They’re designing the system that keeps getting better.

Pete Sena · Digital Surgeons

01 / The scene

I watched one of our designers generate the same screen five different ways with Figma’s agent. Five takes on the canvas, side by side, in minutes.

Then they dropped into a coding tool and vibe-coded a quick prototype of the one they liked. It worked. They looked at it, nodded, and closed the tab.

Everything we learned in that session went with it. The four rejected versions and why they lost. Tomorrow, the same designer starts from scratch.

That’s the old muscle. Research, explore, make, hand off, done. A line with an end. We’ve run on it for twenty years, and it’s what costs us now, because the tools got fast and the line is the slow part.

02 / The shift

A path ends. A loop keeps learning.

A path has a destination. You get there, you stop, you start another path. A loop has a goal and a way of noticing how far off it is. Every lap, it’s a little less wrong.

Here’s the version designers can run today.

01

Set the goal and the constraints

The real problem, the design system, your taste. Written down, not floating in your head.

02

Let the agent generate

Five options or fifty, whatever the problem needs. This is the cheap part now.

03

Sense what’s working

You, or the system, look at the output. Rough, fast, honest. This is where taste earns its keep.

04

Refine the system, not the screen

Fix the constraints, the rules, the skill file, the components. Don’t nudge pixels in one frame and walk away.

05

Run it again

The next generation starts better because the system got better. Then you do it again.

Boris Cherny, who built Claude Code, says he doesn’t prompt Claude anymore. “My job is to write loops.”

Designers are better set up for this than almost anyone. You already think in systems. You already have taste. That’s the scarce part now: execution is getting cheap, judgment isn’t. The loop is where your judgment compounds instead of evaporating.

03 / Right now

What a practical loop looks like today.

EXPLORE

Figma agent

On-canvas, parallel prompting, aware of your design system. That last part is the trick. Point it at the system, not a blank frame.

PROTOTYPE

Vibe coding

Turns a direction into a working prototype fast enough that you test the idea instead of defending it.

COMPOUND

The design system

Tokens, components, variables, rules. Every lap is a run. This is the track getting better.

SHARE

Skills, rules and prompt libraries

Files the whole team edits. One person’s fix becomes everyone’s starting point.

04 / Try it

Run the loop.

Same screen, five passes. Flip between Loop and Path and watch what carries over.

Pass 0 of 5

--on-system score (illustrative)

Rules file

    Pro plan$24Choose planClick here to get started today

    05 / Signals

    Signals that clicked.

    Five pieces that changed how I think about this. Each one opens in a new tab.

    Microsoft Design

    Designing loops, not paths

    Cybernetics turns “human in the loop” from a catchphrase into a practice: a goal, something sensing, a controller deciding, an action. You can be the controller. Start here.

    ↗

    InnoLead · Todd Reily

    The Death of the Design Process (and the Rise of the Infinity Loop)

    The double diamond collapses into one continuous learn-and-build cycle. You bounce between reframing the problem and prototyping many times a day, with no handoff in between.

    ↗

    The Helix Loop · Suhit Anantula

    5Ps of Loop Design

    Perceive, Perform, Portfolio, Pause/Promote, Progress. A loop as an operating system for working with AI: you bring context and judgment, it brings scale.

    ↗

    YouTube · YC Head of Design, Eve Bouffard

    How To Design With AI

    Parallel variations, dense context, and the idea that imagination is the real bottleneck.

    ↗

    YouTube · UI Collective

    Design with AI: Full Guide

    Practical workflows across Claude, Figma, coding agents and design systems.

    ↗

    06 / Compounding

    The system is the asset.

    Screens ship and get forgotten. The system is what you keep. Every lap should leave it a little better than it found it.

    Better tokens and components→Better AI output→More reuse→A stronger system

    01

    The design system

    Tokens, components, variables, rules. Tighten them and the agent’s output tightens with them. That’s the reuse that matters: not copying a frame, but starting every generation closer to right.

    02

    Shared skills, rules and critique habits

    When someone fixes a bad output, the fix goes into the rules file, not just the screen. One person’s correction becomes the whole team’s starting point. Critique stops being a meeting and starts being a commit.

    03

    Product and user feedback

    Keep it simple. What people actually do with the thing goes back into the system as a constraint, a component change, a new rule. Ship it. Learn. Repeat.

    07 / Monday

    Your Monday move.

    Pick one. Not all three.

    A

    Map one workflow as a loop.

    Take something your team does every week. Write down the goal, what generates, how you’ll tell what’s working, and where the fix gets saved. If you can’t name where the fix goes, that’s your gap.

    B

    Strengthen one part of the system.

    Pick a token set, a component, or a rule. Fix it. Then regenerate the same screen and check whether the output is measurably better. If it isn’t, you fixed the wrong thing.

    C

    Run a parallel-variation critique.

    Generate five directions on the same brief. Critique them together, out loud. Whatever the room rejects gets written down as a rule.

    Ask the better question.

    Prototype before you pitch.

    Let AI do the boring parts.

    Ship it. Learn. Repeat.

    Run one lap this week. Then run the next one and see what changed.