# Privacy-Protected Case Study: PIXIE at the AI Skills Jam

**Author and systems architect:** Dominique Devereaux  
**Project:** Loptr Lab / PIXIE Device Stewardship  
**Status:** Public-safe discussion draft

## What this case study includes

A first-person account of using an AI assistant to recover structure from a complicated creative-production project after a difficult human collaboration. The work became a test of accessibility, authorship, provenance, memory support, and responsible escalation to qualified humans.

## What it excludes

- Private correspondence
- The identity of collaborators or family members
- Diagnoses, medications, health measurements, or provider details
- Unpublished third-party creative work
- Claims of endorsement by OpenAI, Apple, or event hosts

## Starting barrier

The creative process required extensive working memory, task switching, file organization, and reconstruction of prior decisions. Those demands were not incidental: they determined whether the creator could continue the work and preserve authorship.

## What the AI assistant helped with

- Recovering project structure from scattered material
- Converting a large goal into smaller, reviewable steps
- Separating documented evidence from interpretation
- Drafting accessibility, privacy, consent, and licensing language
- Keeping public examples synthetic while preserving a path to future research

## Where the AI assistant failed or remained unsafe

- It could begin with incomplete context.
- It could flatten unusual creative intent into generic language.
- It could present a plausible inference too confidently.
- It could fail to ask the clarifying question that a qualified human would recognize.
- Its memory and access boundaries required explicit governance.

## Where qualified humans remain essential

- Legal and intellectual-property advice
- Disability-benefit and tax counseling
- Clinical interpretation and crisis response
- Institutional review, research ethics, and participant protection
- Security and privacy assessment
- Community accountability, consent, and authorship decisions

## Systems insight

Accessibility is not a decorative layer. When a tool externalizes memory, makes state visible, preserves provenance, and reduces unnecessary context switching, it changes whether a disabled creator can participate as an author and systems thinker.

## Current ask

A 30-minute review or warm introduction to an appropriate product, accessibility, human-computer-interaction research, or community team. Any deeper engagement should define scope, compensation, attribution, data rights, and withdrawal terms before private material is shared.