Designing the platform for enterprise
AI orchestration

What we did

— SAAS platform redesign
— Design system
— Marketing website
— Visual language refresh
— Ongoing design partnership

Make distributed AI workflows visible, debuggable, and reliable

Union.ai provides the orchestration layer that powers some of the most demanding AI and ML workloads in production today. Companies like Lyft, Spotify, Toyota, and Lockheed Martin rely on Union to orchestrate distributed pipelines—handling everything from petabyte-scale data processing to latency-sensitive model inference.

As Union prepared to launch v2, they faced a design challenge: how do you create an interface that expresses complex execution semantics, cross-resource dependencies, and real-time system state without overwhelming the engineers who depend on it daily?

Our Approach

Strategy & discovery
We started by understanding the engineering workflows Union supports. Through user research and workflow analysis, we mapped how teams orchestrate distributed pipelines, debug failures, and monitor system health at scale. This revealed where UI complexity was blocking productivity.

Systems-level redesign
We redesigned the information architecture from the ground up—creating clear pathways between monitoring, investigation, and debugging. The new structure lets engineers move from high-level system health to individual task logs without losing context or switching tools.

Embedded partnership
We worked embedded with Union’s product team—designers, PMs, and engineers. Weekly sessions and rapid iteration ensured design decisions reflected real technical constraints and user needs. The result: a v2 launch that shipped on time with a cohesive, scalable design system.

“Good Smile brought real rigor to our product design process, helping us align early, move faster, and ship higher-quality work.

Their design artifacts directly informed our API and implementation decisions, which meant less rework and more confidence we were getting it right the first time.

They also partnered closely with our GTM team to reframe union.ai, clearly connecting our product’s value to what matters most to enterprise buyers.”

— John Votta, Head of Product

SAAS platform

Creating clear information hierarchy

Engineers orchestrating distributed AI pipelines need to monitor thousands of concurrent executions while drilling into individual failures. They need to understand: What’s running? What failed? Why? And what’s the downstream impact?

When a pipeline fails at 3am, engineers need answers fast. We designed an information architecture that moves from thousands of concurrent executions down to individual task logs—preserving context at every level. Engineers can scan system health, identify bottlenecks, and debug failures without getting lost in complexity.

From overview to detail: navigating complexity

Engineers move between different views depending on the task at hand—monitoring overall system health, investigating specific runs, or debugging individual tasks. We designed a seamless progression that takes teams from high-level project lists through execution runs and task details, to workflow visualizations and specifications.

Each view is optimized for its purpose: lists for scanning and filtering, dependency graphs for understanding relationships, and spec views for technical inspection. The UI adapts its information density based on what engineers need to accomplish at each level.

Built for technical clarity at scale

Union’s design system needed to handle edge cases that most products never encounter: workflows with 100,000+ tasks, execution states that span hours or days, and real-time updates across distributed systems.

Every component was designed for extreme information density—virtualized lists handle thousands of rows, status indicators communicate at a glance, and consistent visual language ensures engineers can parse complex states instantly. The system scales from simple pipelines to enterprise-grade orchestration without breaking.

Website

Translating technical depth for multiple audiences

The website needed to speak to two very different audiences: executives evaluating orchestration platforms and engineers assessing technical capabilities. We created a visual system that conveys technical sophistication without overwhelming, using clear information hierarchy and concrete examples to make complex concepts accessible.

The result: a site that educates prospects on what Union does and why it matters—without requiring deep Kubernetes expertise to understand the value.