Four formats, one argument:
make the thinking inspectable.
These are working teaching artifacts, not course descriptions. Each one is a different format because it answers a different question — whether a learner can do the thing, whether a room of decision-makers will fund the thing, or whether a group can diagnose why a change stalled.
Every artifact ships as a single self-contained HTML file: readable by a person, parseable by an agent, and versionable in git. That constraint is itself part of what is being taught.
The formats differ on purpose
Pick by what you want to see. The four formats sit at different points of a delivery arc — from a self-serve unit a learner clicks through alone, to a case a facilitator runs with a room.
Learner clicks through alone. Self-paced, stepped, no facilitator. Success = they can perform the task afterward.
What gets said, and what leaves the room. The content backbone of a live talk — every section pinned to a takeaway for two audience tiers.
One argument, worked through. A short position piece that becomes the source material for slides, workshops and reading groups.
A room argues about a decision. HBS-style: exhibits, no verdict. Built for discussion, not for reading alone.
Open any one
Five artifacts default to English with a ZH-TW toggle. The auth unit stays Traditional Chinese.
Low-friction Auth & Third-party Login
Five stepped screens that reframe passwords as a liability rather than a feature — a toggle exposes the ten things you must own forever if you roll your own, before the unit walks the OAuth authorization-code flow and where credentials may and may not live.
The Token Ledger, Synthetic Users & the Limits of Distillation
A 90-minute talk organised as three management questions — manage cost (a token ledger, and three line items you can cut this week), manage process (synthetic users as a low-budget POC loop), and manage boundaries (when distillation stops being learning and starts being appropriation). Each section closes with a two-tier takeaway: one for people who will build, one for people who will fund.
The Prototype Ladder Grew a Rung
Paper → wireframe → interactive survived the arrival of AI intact. What changed is the middle rung: an exploration stage that now has real output and a borrowed name, Annotated Prototype. Argues the three diving heights — chat, design tool, code — differ by sunk cost, not capability, and shows one audit concept climbing two of them so you can see that fidelity rose barely while convergence rose a lot.
The Demo Landed. The Catch Mechanism Didn’t.
Two cohorts, eleven teams: most reached ADKAR Ability. The shared stall was Reinforcement — who intercepts a wrong output, what gets measured, who owns the miss. Reads the closeout through ADKAR, Kotter, Prosci and IBM’s 2025 AI-change principles, and treats “every group shipped a demo” as the finding, not the win.
Four Weeks to a Runnable Work System
A four-week build, not a reading list: locate the tool chain, audit and variant-test, turn design knowledge into a queryable wiki, then close a POC loop — candidate in, audit run, write-back. The takeaway is a system that runs, not a framework that sits on a slide.
Five Lessons Against the 2026 Market
Maps a five-lesson AI/UX sequence onto what’s already for sale. Lessons 1–3 have decent competitors; 2, 4 and 5 are close to empty — design knowledge as a queryable repository, red-team acceptance that writes back, and workflow governance for design leads.
Co-Creating an AI / UX Book
A two-sided manuscript built with reading-group alumni: Side A walks a practitioner from foundations to applied product decisions; Side B is a strategy path for leaders managing AI-augmented teams. Drafts iterate against real trial-reader reactions.
Anonymization & AI-use disclosure
Client material. The closeout is a teaching artifact from real delivery, published for classroom and coaching discussion only — not as evidence that anyone managed anything well or badly. No commercial terms appear. Read it as a case, not a scorecard.
AI use. Instructional design, argument and structure are mine and I am accountable for them; front-end implementation is AI-assisted, with every technical claim verified line by line. This disclosure follows the transparency template published in the AI/UX practice-flow lecture — a course that asks learners to own their AI output should hold its own materials to the same rule.