←

→

Building an AI‑Native Design Organization

At Kustomer, I rebuilt the design operating system around a new reality: designers, engineers, and AI systems working in the same loop.

Company

Kustomer.com

Role

Head of Product Design

Focus

AI‑native product development, operating systems, and team adoption

Work

Impact

Thought Currents

About

Resume ↗

Contact me

Kustomer · Case 01 · Organizational impact

Next: Enterprise AI Agents for Customer Ops

←

→

Building an AI‑Native Design Organization

At Kustomer, I rebuilt the design operating system around a new reality: designers, engineers, and AI systems working in the same loop.

Company

Kustomer.com

Role

Head of Product Design

Focus

AI‑native product development, operating systems, and team adoption

Kustomer · Case 01 · Organizational impact

Next: Enterprise AI Agents for Customer Ops

←

→

Building an AI‑Native Design Organization

At Kustomer, I rebuilt the design operating system around a new reality: designers, engineers, and AI systems working in the same loop.

Company

Kustomer.com

Role

Head of Product Design

Focus

AI‑native product development, operating systems, and team adoption

3d → 3 hrs

Concept → working prototype

↓80%

Time to testable direction

75%

Static specs replaced

3d → 3 hrs

Concept → working prototype

↓80%

Time to testable direction

75%

Static specs replaced

3d → 3 hrs

Concept → working prototype

↓80%

Time to testable direction

75%

Static specs replaced

00 Brief

Specs loosely interpreted, rework silently absorbed, AI used in isolation.

The brief: rebuild how design, engineering, and AI work together so a design question becomes a working, evaluated prototype in hours, with quality standards the whole team shares.

00 Brief

Specs loosely interpreted, rework silently absorbed, AI used in isolation.

The brief: rebuild how design, engineering, and AI work together so a design question becomes a working, evaluated prototype in hours, with quality standards the whole team shares.

00 Brief

Specs loosely interpreted, rework silently absorbed, AI used in isolation.

The brief: rebuild how design, engineering, and AI work together so a design question becomes a working, evaluated prototype in hours, with quality standards the whole team shares.

01 AI‑native principles

Work in live systems over static specs

Design prompts, outputs, rules, and recovery paths, not just flows.

Evaluate AI outputs against real use cases.

Co-create with engineering from the start.

Use real data and real user signals as early as possible.

01 AI‑native principles

Work in live systems over static specs

Design prompts, outputs, rules, and recovery paths, not just flows.

Evaluate AI outputs against real use cases.

Co-create with engineering from the start.

Use real data and real user signals as early as possible.

01 AI‑native principles

Work in live systems over static specs

Design prompts, outputs, rules, and recovery paths, not just flows.

Evaluate AI outputs against real use cases.

Co-create with engineering from the start.

Use real data and real user signals as early as possible.

02 Player‑coach adoption

I rewired my own practice first, then used that learning to help the team move from static design artifacts to AI-assisted prototypes, research synthesis, product critique, and shared evaluation systems.

Toolkit

Cursor

Claude Code/Design

Code-based prototyping

AI-assisted research synthesis and product analysis

designmd.app

aiuxplayground.com

ChatGPT

02 Player‑coach adoption

Toolkit

Cursor

Claude Code/Design

Code-based prototyping

AI-assisted research synthesis and product analysis

designmd.app

aiuxplayground.com

ChatGPT

02 Player‑coach adoption

Toolkit

Cursor

Claude Code/Design

Code-based prototyping

AI-assisted research synthesis and product analysis

designmd.app

aiuxplayground.com

ChatGPT

03 Operating model

Four layers. Each one makes the next cheaper to enforce.

Guidance is coded, then built into the system, then checked by machine, then flagged if human review is needed. Everything below is a real file in the design repo.

01GUIDE

design-rules.mdGUIDE 01

Codify Design Judgment

Answers: What is right?

Turn recurring design decisions into reusable instructions for humans and AI.

## Excerpt · rules 07, 19, 24, 31

R-07Destructive actions state the blast radius before confirming.

R-19Error copy states what happened, then what to do. Never “Something went wrong.”

R-24Tables over 50 rows ship with filter and column sort.

R-31AI suggestions stay visually distinct until the user accepts them.

## Output

Version-controlled guidance that gives designers, engineers, and AI a shared baseline.

This layer guides behavior. It does not guarantee compliance.

02CONSTRAIN

design-system/CONSTRAIN 02

Make the Right Thing Easier

Answers: How do we make right the default?

Move repeatable design decisions out of prompts and into the system itself.

## What we encode

Semantic design tokens · Typed component APIs · Approved variants · Accessibility defaults

## Output

A choice like primary, instead of raw CSS.

GOOD<Button variant="primary" size="medium" />

NOTbackground: #4263EB; padding: 12px 18px; border-radius: 6px

03VERIFY

flow-audit.mdVERIFY 03

Test Before Anyone Reviews

Answers: What slipped through?

AI audits every prototype against the rules before a human sees it.

## Audit output · /tools/edit · v1

Fail: R-07Destructive action lacks blast radius

Found“Are you sure?”

Fix“Update this tool? 5 procedures will be affected.”

Fail: R-19Generic error copy ×3

Found“Something went wrong”

Fix“Couldn’t save the tool. Check the required fields and try again.”

Pass: R-02Contrast ≥ 4.5:1 — 48 / 48 text nodes

Pass: R-31AI suggestion styled distinct — 12 / 12

## Output

A pass / fail list attached to the PR. Nothing merges with an open ✗.

04GOVERN

governance.mdGOVERN 04

Keep Judgment Human

Answers: What may AI decide?

Define where automation ends, exceptions are reviewed, and the system learns.

## Decision boundaries

Reversible / Low stakes → AI acts, system logs

Reversible / High stakes → AI proposes, human confirms

Irreversible → Human decides, AI assists

## Learning loop

Exception discovered → Review → Codify → Test → Reuse

01GUIDE

design-rules.mdGUIDE 01

Codify Design Judgment

Answers: What is right?

Turn recurring design decisions into reusable instructions for humans and AI.

## Excerpt · rules 07, 19, 24, 31

R-07Destructive actions state the blast radius before confirming.

R-19Error copy states what happened, then what to do. Never “Something went wrong.”

R-24Tables over 50 rows ship with filter and column sort.

R-31AI suggestions stay visually distinct until the user accepts them.

## Output

Version-controlled guidance that gives designers, engineers, and AI a shared baseline.

This layer guides behavior. It does not guarantee compliance.

02CONSTRAIN

design-system/CONSTRAIN 02

Make the Right Thing Easier

Answers: How do we make right the default?

Move repeatable design decisions out of prompts and into the system itself.

## What we encode

Semantic design tokens · Typed component APIs · Approved variants · Accessibility defaults

## Output

A choice like primary, instead of raw CSS.

GOOD<Button variant="primary" size="medium" />

NOTbackground: #4263EB; padding: 12px 18px; border-radius: 6px

03VERIFY

flow-audit.mdVERIFY 03

Test Before Anyone Reviews

Answers: What slipped through?

AI audits every prototype against the rules before a human sees it.

## Audit output · /tools/edit · v1

Fail: R-07Destructive action lacks blast radius

Found“Are you sure?”

Fix“Update this tool? 5 procedures will be affected.”

Fail: R-19Generic error copy ×3

Found“Something went wrong”

Fix“Couldn’t save the tool. Check the required fields and try again.”

Pass: R-02Contrast ≥ 4.5:1 — 48 / 48 text nodes

Pass: R-31AI suggestion styled distinct — 12 / 12

## Output

A pass / fail list attached to the PR. Nothing merges with an open ✗.

04GOVERN

governance.mdGOVERN 04

Keep Judgment Human

Answers: What may AI decide?

Define where automation ends, exceptions are reviewed, and the system learns.

## Decision boundaries

Reversible / Low stakes → AI acts, system logs

Reversible / High stakes → AI proposes, human confirms

Irreversible → Human decides, AI assists

## Learning loop

Exception discovered → Review → Codify → Test → Reuse

From direction to quality

05 What I built

Build 01

text

Moved design team from static handoffs to AI-native co-creation: prototype with LLMs, test real outputs, iterate in hours.

Build 02

text

Introduced Markdown-based skills to standardize AI usage across prompts, component rules, output evals, and human-in-the-loop guardrails.

Build 03

text

Moved design critique from subjective review to observable output quality. AI-generated work is evaluated against real use cases, defined edge cases, and documented product intent before shipping.

Build 04

text

Eliminated traditional handoff by embedding design and engineering in the same loop: co-defining behavior, prototyping together, and reviewing outputs collaboratively.

05 What I built

Build 01

text

Moved design team from static handoffs to AI-native co-creation: prototype with LLMs, test real outputs, iterate in hours.

Build 02

text

Introduced Markdown-based skills to standardize AI usage across prompts, component rules, output evals, and human-in-the-loop guardrails.

Build 03

text

Moved design critique from subjective review to observable output quality. AI-generated work is evaluated against real use cases, defined edge cases, and documented product intent before shipping.

Build 04

text

Eliminated traditional handoff by embedding design and engineering in the same loop: co-defining behavior, prototyping together, and reviewing outputs collaboratively.

05 What I built

Build 01

text

Moved design team from static handoffs to AI-native co-creation: prototype with LLMs, test real outputs, iterate in hours.

Build 02

text

Introduced Markdown-based skills to standardize AI usage across prompts, component rules, output evals, and human-in-the-loop guardrails.

Build 03

text

Moved design critique from subjective review to observable output quality. AI-generated work is evaluated against real use cases, defined edge cases, and documented product intent before shipping.

Build 04

text

Eliminated traditional handoff by embedding design and engineering in the same loop: co-defining behavior, prototyping together, and reviewing outputs collaboratively.

06 Impact

Compressed idea-to-working-output timelines from days or weeks to hours, reduced interpretation gaps and rework, and tested design quality against real conditions earlier.

06 Impact

Compressed idea-to-working-output timelines from days or weeks to hours, reduced interpretation gaps and rework, and tested design quality against real conditions earlier.

06 Impact

Compressed idea-to-working-output timelines from days or weeks to hours, reduced interpretation gaps and rework, and tested design quality against real conditions earlier.

07 Key insights

Design thinking does not disappear. It moves from static files into system behavior: prompts, rules, evaluation, and the user’s voice in the process.

07 Key insights

Design thinking does not disappear. It moves from static files into system behavior: prompts, rules, evaluation, and the user’s voice in the process.

07 Key insights

Design thinking does not disappear. It moves from static files into system behavior: prompts, rules, evaluation, and the user’s voice in the process.

★ What colleagues say ★

Yvonne is a cheetah-mounted hot pink laserbeam.

At Kustomer she killed the handoff. Not metaphorically. She moved the entire design org out of static specs and into AI-native co-creation with engineering: prototyping directly in IDEs, testing outputs against real data, shipping in hours instead of days.

She codified the whole thing into reusable markdown-based skills so AI stopped being individual experimentation and became a shared operating system. Then she moved critique from vibes to observable output quality. Then she cut time to decision by 80% because why not

Arkady Sokolov
Product, AI | ClickUp

★ What colleagues say ★

Yvonne is a cheetah-mounted hot pink laserbeam.

At Kustomer she killed the handoff. Not metaphorically. She moved the entire design org out of static specs and into AI-native co-creation with engineering: prototyping directly in IDEs, testing outputs against real data, shipping in hours instead of days.

Arkady Sokolov
Product, AI | ClickUp