Project Name
Project Overview
Role
Designing Harry : An Agentic AI Companion
How do you design AI that does things FOR students, not just suggests things TO students?
For my Advanced User Design: Agentic AI Studio class, I designed the behavioral architecture for Harry - an AI agent that reduces cognitive load for college students by making decisions on their behalf. Unlike traditional task management apps, Harry doesn't just organize information; it autonomously extracts tasks from Canvas, detects stress patterns, and surfaces resources when needed.
AI Experience Designer
Duration
10 weeks
The Problem -
-
AI in healthcare risks public backlash when ethical concerns aren't addressed
-
Companies struggle to predict what will trigger moral outrage
-
Two critical scenarios: AI-generated health content & patient data usage
My Research Process -
-
Applied Haidt's Six Moral Foundations Theory to predict user reactions
-
Mapped each foundation to AI healthcare scenarios
-
Developed ethical UX framework with actionable design guidance
The Challenge - Decision Fatigue, Not Lack of Tools
The Real Problem
College students interact with 5+ platforms daily :
-
Canvas for assignments
-
Google Calendar for scheduling
-
Financial aid portals
-
Advising systems
-
Mental health apps
Key Research Insight
After conducting primary research with students it said showed students spend 40% of their time deciding WHAT to do next. They don't need more tools they need an AI agent that eliminates decisions by doing things on their behalf.
From Assistant to Agent: The Design Shift
Most AI assistants wait to be asked. Harry observes, predicts, and acts with user permission.
Reactive
Waits for a command
"SIRI"
Suggestive
Offers options
"ChatGPT"
Proactive
Initiates unprompted
"Duolingo"
Agentic
Takes action automatically
"Harry the AI"
Three key autonomous behaviours that eliminate cognitive load
How Harry Reduces Decisions
📚
Automatic Task Extraction
ZERO MANUAL ENTRY
-
Syncs with Canvas automatically
-
Prioritizes by deadline + context
-
Adds to dashboard
-
Suggests optimal work time
Result: 5 decisions eliminated
🔍
Predictive Pattern Detection
AUTONOMOUS INSIGHTS
-
Analyzes mood over time
-
Detects stress patterns
-
Diagnoses causes
-
Proposes solutions
Result: 7 decisions automated
🎯
Goal-Driven Resource Surfacing
ACTIVE DELIVERY
-
Monitors goals passively
-
Matches opportunities
-
Surfaces at right moment
-
No browsing required
Result: From passive to active
Before
10 Steps, 5 App Switches, Cognitive Overload - Every task requires multiple manual decisions and context switches (shown by low fidelity wireframes)
Check Canvas
Read Details
Decide Priority
Switch Apps + Write Down
Switch Apps + Add Deadline
Set Reminder + Plan
After
1 Glance, 0 Manual Steps, Complete Context - Harry syncs Canvas, prioritizes, and surfaces everything automatically
The Interface of the AI
The Interface of the AI
