WorkAI Refinement WorkflowValidated

AI interfaces make generation easy, but refinement remains difficult.

Role: Lead Designer
Team: 3 designers on refinement use case
Timeline: August 2025 to January 2026
Tools: Figma · Maze · Usability testing · Medium

Overview

As part of Carnegie Mellon University’s UI for AI initiative, we designed interaction patterns that give people more control over refining AI-generated ideas instead of repeatedly starting over.

My role

I led a team of three designers exploring how designers and writers refine AI-generated work, guiding our research, design explorations, and prototyping.

Problem

Users shouldn’t have to choose between restarting and manual editing.

AI tools generate quickly, but refinement often breaks down after the first draft. Users are left choosing between regenerating everything or editing manually.

How might we support refinement across multiple levels while preserving structure and user intent?

Research

Research showed users needed refinement they could see and control.

I analyzed existing AI tools and refinement workflows to understand how users maintain control while iterating on generated content. Existing systems tended to prioritize either broad exploration or precise edits.

Adobe Firefly Generative Fill selecting a region and generating content in place — Localized editing
Users could refine one area without changing the rest. Inspired by Adobe Firefly / Generative Fill.

SpecifyUI Region SPEC Split highlighting structured UI regions on a dashboard — Structured refinement
Users could refine content at multiple levels. Inspired by SpecifyUI.

01Support multi-level refinement
Users naturally switched between broad revisions and precise edits.
02Preserve structure during refinement
Participants wanted to improve selected content without losing surrounding work.
03Keep users in control
People needed clearer visibility into what AI changed before accepting edits.

Users rarely accepted first outputs.

“Show me what changed so I can make the call myself.”
Participant, usability testing

Concept testing

Testing refinement interactions pointed toward one clear workflow

We tested mid-fidelity prototypes with users through interviews and task-based walkthroughs focused on refinement workflows. Participants completed tasks across broad edits, detailed editing, merge workflows, and version history interactions while sharing where friction occurred.

My concept was selected because it supported both broad and precise refinement without interrupting workflow.

Wireframe showing response options, highlighted phrase, tone menu, and refined output — Overall refinement
Broader changes like rewrite, merge, tone, or structure.

Wireframe showing inline text selection and a rewrite prompt for targeted refinement — Inline selection
Highlight exactly what needs refinement.

Wireframe showing inline text selection, a More Formal refinement menu, and the refined output snippet — Inline refinement
Refine selected content without regenerating everything.

Decisions

Three design decisions made AI refinement more controllable.

Instead of optimizing for better first outputs, we explored how users could refine AI-generated content while preserving structure, intent, and momentum.

Designed refinement at multiple levels

Users shifted between revising entire responses and fine-tuning individual words.

Replace regeneration with targeted refinement.

“It rewrote my intro without asking.”
Participant, usability testing

Try again control that regenerates the full AI output — Before · Whole output regeneration
Regenerating full outputs often overwrote parts users wanted to keep.

Inline text selection with Ask AI and refinement options on highlighted text — After · Selected-region refinement
Enabled targeted editing so users could refine selected regions without changing the rest.

Version visibility made refinement easier to evaluate

Added side-by-side version comparison and navigation.

Side-by-side comparison of Current Response and Refinement V1 with version preference controls — Users could judge AI changes before committing to them.

Solution

Supporting refinement without restarting

We designed a refinement workflow that combined broad edits, targeted changes, and version comparison so users could steer AI outputs without losing context or starting over.

Overall Refinement. Broad changes without rewriting prompts.

Detailed Editing. Refine specific regions without changing everything else.

Version Comparison. Review iterations before committing to changes.

Prototype testing

Testing validated the refinement workflow.

Maze testing validated that users understood the refinement model and successfully completed tasks across overall refinement, inline refinement, and comparison workflows.

Key takeaway: Users preferred inline refinement. Participants consistently preferred localized edits over repeated prompting.

100%
Task completion across core refinement tasks
Users completed overall refinement, inline editing, and comparison tasks successfully.
0%
Misclick rate during refinement tasks
No navigation or shortcut confusion appeared during testing.