AI in the Design Process

Turning design friction into forward momentum with AI

AI in the Design Process

Overview

AI-assisted prototyping revealed the team's real bottleneck and unlocked a better approach for the team.

Design was perceived as too slow. Rather than optimize for speed, I questioned this assumption and designed an experiment with AI-coded prototypes to test it. When teams got access to faster prototyping, they didn't celebrate speed revealing they wanted autonomy. This insight led me to recommend the team shift towards the build-to-learn model, a far greater contribution than making design faster.

The Problem

Design was seen as the blocker to delivering on new features.

When I joined to replace a two-person team, I inherited a reputation that Design was too slow. As the sole designer, I built infrastructure to help move and scale faster. Output improved with not only speed but greater consistency. Yet the perception remained the same.

Design was seen as the blocker to delivering on new features.

Hypothesis

Faster design should have resolved the friction if speed were really the constraint.

I had a hunch speed wasn't the real ask. In order to find out what was behind the tension I sensed, I came up with an experiment to test if improving speed once more would alter the perception of Design in the team culture.

Experiment

AI-coded prototypes showed potential for speeding up the most time-intensive parts of the process.

I made coding accessible in a way it hadn't been before. Combined with our component library, it meant I could iterate on complex interactions much faster than traditional design tools allowed. This was the moment to test what that capability could unlock for the process itself.

Complex interactions

Code offers a way to create more complex prototypes much faster than can be done in Figma.

Faster ideation

Being able to create a prototype much faster allows for rounds of ideas to be worked through more quickly.

Scalable prototyping

Coded prototypes can be made to be cumulative, with each one becoming the foundation for the next.

Speed up development

Coded prototypes using the product's tech stack have potential to speed up handoff.

01

No-code tools were the most accessible entry point, and, as it turns out, the most limited

No-code tools promised speed but hit their ceiling fast. Local environment coding tools such as Cursor offered more flexibility at the cost of a steeper learning curve. Scalability meant choosing the harder path.

Levels of AI coding tools

Levels of AI coding tools

02

AI assistance made the learning curve manageable.

With no coding background, I leaned on AI assistance to work through the basics and build up my foundations.

Cursor AI-assisted Coding

Cursor AI-assisted Coding

03

Starting from zero, I rebuilt both products in code and built the testing infrastructure around them

No access to the sandbox environment opened another opportunity to build scalable prototypes. I built the mock version of our claimant platform and our claims professional portal, starting with the applicable flows from live projects.

Claim Professional's portal prototype

Component library built in Storybook

Component library built in Storybook

Results

New process excelled, and proved "speed" wasn't the real ask.

Days, not weeks.

Complex interactions that would have taken an additional week in Figma were resolved in days, with clearer outcomes and faster creative clarity.

Faster, realer feedback.

Handing off working prototypes for internal testing meant stakeholders reacted to something real, and feedback came back faster.

Leadership wanted in.

The Head of AI and Head of Product didn't just approve the process. They brought their own ideas to it and requested access themselves.

What surprised me was the immediate request from Heads of departments for access while the feedback stayed the same.

Design is still too slow.

It confirmed the real constraint wasn't Design's speed. It was the desire to build their own ideas out.

Analysis

Crosstie's internal tension reflected a shift already reshaping how product teams were changing across the industry

Crosstie's internal friction wasn't unique. It was a microcosm of a larger shift reshaping product teams everywhere. With AI lowering barriers to research, strategy, design and even code, the competencies once exclusive to Product, Design, and Engineering have become more permeable. Yet this blurring doesn't mean roles collapse into one; rather, it's about who wants to own which competencies, and what your organization actually needs.

Product Trio roles diverging with AI lowering barriers

Product Trio roles diverging with AI lowering barriers

For Crosstie specifically, a build-to-learn model made sense:

  • 1. Established design system and a library of components to pull from that would cover at least 90% of their design needs.
  • 2. If current trajectory maintained, they could achieve their $10-12M ARR goals within 1 to 2 years with the given roadmap and very light Designer input.
  • 3. The team had proven they could whip up technical prototypes at a speed which made build-to-learn less risky.
  • 4. Heads of departments already showed a desire to bring their own ideas to life. If the Design function stayed as it was currently set up, it would continue to create friction.

But the real insight was recognizing that as building becomes easier, the needs shift from efficiency to focus. Without design thinking to guide which experiments matter most, teams risk scattering their energy across too many directions.

The companies that'll win are those that keep design thinking as their north star, even as design's output becomes more democratized.

Findings

A recommendation that seemed to argue against design was the one that proved its value most clearly

I recommended the team adopt a build-to-learn model, shifting toward fractional design support. It was adopted immediately. The design role transformed from gatekeeper to guide, present for strategic moments that truly needed design thinking, while the team moved fast on their own.

It felt counterintuitive to recommend Design step back, but that constraint was exactly what unlocked the team's potential. The team could move fast, and Design stayed where it added real value.