Eden AI

The tension

Eden AI is a full-stack platform where developers build and validate AI-powered workflows. Getting started meant picking one of three separate doors in: a template library, an AI assistant, or a blank canvas, each living in its own interface, with build, test, and configuration split across still more screens. The variety was meant to read as flexibility. In practice it just meant more places to get lost before doing any real work. Only 6% of people who opened the builder ever finished a workflow.

What I designed

Instead of improving each entry point on its own, I replaced all three with one: a single search component that acts as the entry point for every workflow-creation action. A tab press switches it between browsing templates and invoking AI assistance, so the same field carries both without asking anyone to think about which mode they’re in.

• Consistency: one interaction model, standardised preview cards, and shared keyboard shortcuts across surfaces that used to behave differently from each other.
• Progressive disclosure: advanced capability surfaces as someone explores, instead of front-loading every option on day one.
• Context-aware behaviour: the component reads what a person is doing inside the builder and adapts its suggestions and ordering accordingly.

The workflow builder itself moved from a cramped vertical layout to a horizontal one, giving the canvas the room the work needed, with preview cards and documentation surfaced inline instead of parked in a separate reference tab.

Where it stands

Every decision traced back to user feedback sessions, checked against analytics and session recordings rather than assumption. Context switching and cognitive load went down; more of the platform’s capability became discoverable through categorised results instead of buried menus; and the builder held up for people across a wider range of technical skill, not just the developers who already knew where everything was.

Studio write-up at Tcules