Oversight AI Agent
Overnight job failures, turned into ready-to-approve tickets — designing an agentic triage system for enterprise batch job failures, from stakeholder interviews through to a fully prototyped human-in-loop approval experience.
ROLE
Lead UX · AI Product Design
PLATFORM
Web
DURATION
6 months
STATUS
Shipped
AI agents
Human-in-loop
Enterprise UX
Workflow design
Agentic prototyping
99%
Of errors are repeat failures across 16,000+ logs
0→1
Full agentic product designed and shipped
24h
From overnight failure to resolved ticket
01 — THE PROBLEM
Every night, 10+ automated jobs move data across enterprise systems. When one fails between 2–10 AM, a person must step in before anything moves. They answer the same questions from scratch every time: whose job is this, how serious is it, has this happened before? Then they write the support ticket by hand.
12 years of error logs told a clearer story: 16,000+ real errors, 99% repeats. The challenge: how do you give operations teams consistent, fast, auditable triage without removing human judgement from the final decision?
05 — WHAT I LEARNED
Agentic UX requires a new design contract: the AI must earn the right to act autonomously by first proving its reasoning. Every auto-generated ticket includes a confidence score, a pattern match from history, and a one-tap override. Designing that audit trail was what made stakeholders comfortable moving from human-first to AI-first triage.
NEXT PROJECT
GearUp — Connected car app for families
View