SaaS · Agentic AI
2025
Pulse Agent Console
Client
Pulse (concept)
Role
Senior Product Designer
Timeline
12 weeks
Tools
Figma, Claude, Framer
Designing how operations teams supervise AI agents: what they did, why they did it, and when a human should step in.
01
Problem
Teams trusted the agents for simple tasks but had no way to audit decisions. When something went wrong, nobody could tell which step failed or why.
02
Solution
A run timeline that shows every agent step with its reasoning, confidence and the data it used, plus clear hand-off points for human approval.
03
Process & Workaround
We prototyped three mental models with ops leads: chat log, task board and timeline. The timeline won because it answered ‘what happened?’ fastest. A lightweight approval queue covered the risky steps.
04
User Journey
The journey moved from setup, to first run, to first failure. Designing for the first failure turned out to be the moment that built long-term trust.
05
Wireframe → Visual
Wireframes focused on information density without noise. The visual design used monochrome for system output and a single accent for anything that needs a human.
(A) Wireframe
(B) Visual design
06
Prototype
An interactive prototype simulated live agent runs, including failures, so testers could practise recovering from errors.
07
Outcome
Pilot teams approved more automation once they could see inside each run, and time spent investigating failures fell significantly.
+64%
Tasks automated
−47%
Time to resolve failures
92%
Trust score in pilot
08
Next Project