Agent UXB2B SaaSWorkflow
2026 / AI Product Design Lead / 0-to-1 Product Definition
Enterprise Agent Copilot
A controllable, auditable, and explainable agent workspace for enterprise operations teams.
+38%Task completion
-31%Takeover time
12Core flows
PDF
Case deck PDF
Problem framing, tradeoffs, and launch results.
Video
Agent takeover walkthrough
Exception, confirmation, and human takeover flow.
Images
Key interface gallery
Workspace, task states, and audit trail.
01
Background
The operations team wanted agents to execute repetitive work, but the early experience was only a chat box without task state, source references, risk controls, or human takeover.
02
My role
- Defined the agent workspace IA and task state model.
- Designed references, confirmations, undo, exception takeover, and audit trails.
- Aligned product, ML, and engineering on MVP scope and evaluation metrics.
03
Design process
- Interviewed operators and mapped high-frequency tasks with human judgment points.
- Validated controllability and recovery paths through task-flow prototypes.
- Built a state component library for waiting, running, risky, and completed states.
04
Outcome
- Delivered an end-to-end agent workspace, high-fidelity prototype, and design system specs.
- Created a traceable path from model output to human review.
05
Results
- Core task completion increased by 38%.
- Average manual takeover time decreased by 31%.
- Identified 17 high-risk interaction issues before launch.