We design the rollout sequence, the governance model, and the interfaces people actually keep using — so an agentic workflow becomes operating practice instead of a slide.
AIPET — Agency, Inclusiveness, Privacy, Explainability, Trust. Each one derived outward through four layers: governance duty, human-AI philosophy, interaction norms, and cross-agent patterns.
VarBridge is a practicum prototype for hemato-oncology variant interpretation. It gathers the evidence around the request itself, keeps every source's provenance visible, and leaves a human decision wherever an automated score or summary could change the result.
Evidence is scattered. The request gets lost. Scores look final.
Staff open gnomAD, ClinVar, CKB, COSMIC, OncoKB, CIViC, in-silico predictors and the literature, then write the report the physician asked for. The workbench puts those sources in one matrix and records who decided what. It uses no patient data and does not sign out a result. A design concept, not a validated clinical tool.
Open the workbench ↗An HCI and algorithm-auditing track at McGill's ACT Lab, supervised by Dr. Karyn Moffatt — agent transparency, trust reasoning, and what a regulator can actually verify from the public record.
The programme counted 17,415 visits to show that doctors saved time. It asked six patients what it felt like.
Patients were promised they would be told, asked for consent, and able to check or correct what was written. This reads the public record to see what it actually shows — and publishes the register it stands on, so the reading can be checked rather than believed.
Read the case study ↗PrivacyUX Consulting Ltd. is a Montréal practice, registered in British Columbia, working with enterprises that need AI to survive contact with daily operations.
It is led by Joshua Hsiao. Before consulting, he spent five years at Alibaba — Principal UX Architect (P9) for Alimama, the group's marketing and advertising platform, where he built its first production AI copywriting system, and founder of Alibaba Design University.
Earlier: product management at Yahoo!, design research at ASUS, and UX leadership at Taobao, where he grew the research organisation from 15 to 35 people. He now researches agent transparency at McGill's ACT Lab and writes two columns on Agent governance and AI/UX adoption.