ORIGIN

Manhattan Design System (MDS) is the design system powering JPMorgan Chase's consumer products — Chase mobile, Chase web, Chase Pay, Chase Travel. The mandate: rapidly design and deliver digital experiences at the scale of America's largest consumer bank.

PROBLEM SOLVED

Each design team was forking their own system. Stale, desynchronized patterns. The same Chase button appeared in five different shades across production. The design system was tolerated, not loved.

My goal: make the design system loveable. The path: stop telling designers to come to the system. Bring the system to them — inside Sketch and Figma, automatically.

THE ORG I WORKED ACROSS

Before MDS could be loveable, it had to be adopted.

And adoption was failing — not for lack of a system, but because the system fought the designers' workflow.

Same features appearing differently across the application in production.

I conducted an internal audit across product squads to understand the complexity, scale, and bottlenecks of design at JPMC. Three findings showed up repeatedly.

• FINDING 01
  The design system disrupted designers' normal workflow. Adopting MDS meant switching libraries, redoing setup, relearning where things lived. Designers chose the path of least resistance — and that wasn't MDS.
• FINDING 02
  The design org had no measure of quality for design artifacts. There was no shared definition of "MDS-compliant." Drift went unnoticed until QA, or worse, production.
• FINDING 03
  Many designers found it tedious to switch to a new design system. Even motivated designers gave up halfway. The cost of switching exceeded the perceived benefit.

If designers wouldn't come to the design system, the design system would come to them.

Inside Sketch and Figma, right inside the workflow they already used. Drawing parallels from coding: developers had ESLint. Designers had nothing.

THE BIG IDEA

"What if MDS shipped as a linter inside Sketch and Figma — a companion tool that brought the design system to the designer, not the other way around?"

Coding Linter

The MDS Assistant Storyboard

Design Assistant didn't ship loveable. V1 shipped as an "annoying quibbler."

V2 was the version designers fell in love with. The difference between them is the entire lesson of this case study.

VERSION 1 — The linter

Brought MDS fundamental styles into the designer's workflow. Checked ADA contrast, colors, text, borders. Grouped defects by type for batch fixing.

MDS Assistant V1 annoyed designers with mandatory fixes ranging in 1000's sometimes.

• V1 PROBLEM
  Required all errors to be fixed for quality reporting — even on WIP artboards and exploratory designs.
• V1 PROBLEM
  Flagged exploratory work designers knew wasn't done — and made them feel surveilled.
• V1 PROBLEM
  Grouped defects by type, assuming designers wanted to batch-fix. They didn't.
• V1 RESULT
  Did not drive adoption. Designers tolerated it. They didn't love it.

The surprising discovery from alpha research.

Designers were uncomfortable with my grouping structure. They wanted to see defects the way they saw their work — as a layer tree, just like Sketch and Figma.

I changed course. Redesigned Design Assistant's information architecture to mirror the designers' mental model — the same layer tree they already knew. The pivot saved the product.

The MDS Assistant V2

VERSION 2 — The friendly companion

From "annoying quibbler" to friendly companion. V2 stopped policing and started helping.

• V2 FEATURE
  Manages libraries from the design system automatically.
• V2 FEATURE
  Applies correct branding automatically.
• V2 FEATURE
  Creates specifications for engineers automatically — handoff docs generated in seconds.
• V2 FEATURE
  Brings MDS documentation right into Sketch and Figma.
• V2 FEATURE
  Rewires MDS components from Sketch to Figma during the company-wide migration.
• V2 FEATURE
  Still removes defects — but now opt-in, layer-by-layer, designer-led.

Design Assistant v2 : The friendly companion UI with branding tools, library management, and the layer-tree defect view.

Design Assistant is now pre-installed on every JPMorgan Chase designer's machine. MDS engages directly with 5+ brands, 20+ product groups, 200+ designers, and 500+ engineers daily — and the system has grown beyond my team.

What changed in the org.

• 01
  Design Assistant is mandatory and pre-installed on every JPMC designer's machine — switching from MDS is no longer the path of least resistance.
• 02
  MDS expanded beyond my team. Other product squads now contribute components and patterns back into the system.
• 03
  Single source of truth. The org is now actively working to bring the entire JPMC digital experience under one design system.
• 04
  No more rogue brand creation. Once Design Assistant existed, it was easier to use MDS than to fork.

• 01
  Automation should partner, not overtake. V1 tried to take control of the designer's workflow. V2 gave it back. Every automation decision is a power-handoff decision.
• 02
  A linter that says "no" gets ignored. A companion that says "let me help" gets adopted. Same defect-detection engine underneath. Completely different reception. The difference was tone, framing, and whether the tool started with help or accusation.
• 03
  Listen to the alpha. Pivot to the user's mental model. I had a strong opinion on how defects should be organized. Alpha users had a stronger one. The IA pivot was the single highest-leverage decision in the whole project.
• 04
  Make the right way the easy way. Pre-installation, library management, branding automation — every "right thing" became a thing the designer didn't have to do. Compliance through delight, not enforcement.
• 05
  Cross-functional adoption is its own design problem. Designing a design system at the scale of 200 designers and 500 engineers isn't really about components. It's about adoption, incentives, and the politics of standardization.

• 01
  V1 → V2 is the whole product lesson. V1 shipped as a quibbler. The same engine, reframed as a companion, became V2 — and adoption went from "tolerated" to mandatory. Same code, different product strategy. The product is more than the features.
• 02
  Alpha research kills your assumptions. Listen to it. I had a strong opinion on how defects should be grouped (by type). Alpha users had a stronger one (by layer tree). The IA pivot was the single highest-leverage product decision in the whole project.
• 03
  Pre-installation is a distribution strategy. Once we made Design Assistant pre-installed via the corporate Sketch/Figma image, adoption stopped being a product problem and became a default. PMs should own distribution, not just features.
• 04
  Compliance through delight beats compliance through mandate. Pre-installation could have felt forced. Instead V2 wrapped the rules in genuine utility — library management, branding automation, engineer handoff specs. Designers wanted it.
• 05
  Internal SaaS is real SaaS — with weirder distribution. Same playbook applies: customer research, MVP, V1→V2 pivot, growth, retention, NPS. Just no marketing site. The product still has to earn its place every day.