Before buying AI training, ask to see an exercise
Assess a training proposal through its exercise, assessment criteria, preparation and support.
Read article : Before buying AI training, ask to see an exerciseWriting
Putting AI to work.
Learning how to teach it.
Lessons from corporate training, course design and the work I do with AI every day.
AI at work
Choosing training, leading teams and turning learning into everyday practice.
What SMEs Actually Ask About AI
I sat on a panel at an HKPC expo for small businesses. The prepared answers were fine. The audience's questions were better — here are the actual answers.
Read article : What SMEs Actually Ask About AIOne Jewellery Brand, Two AI Trainings in Two Days: Design, Then Operations
A Hong Kong jewellery retailer booked one AI session for its design team, then booked a second for operations the next day. What the two curricula looked like, and what cross-department demand actually signals.
Read article : One Jewellery Brand, Two AI Trainings in Two Days: Design, Then OperationsMicrosoft Copilot Training for Hong Kong Companies: Count the Licenses First
Your company already pays for Microsoft 365, so it assumes it has Copilot. Count the licenses first, then train on the four apps where Copilot earns its seat — and the 6-week structure that makes it stick.
Read article : Microsoft Copilot Training for Hong Kong Companies: Count the Licenses FirstWhy AI Training Doesn't Stick
Most corporate AI training produces zero lasting behavior change. After training 10,000+ professionals across 70+ organizations, here's what actually makes AI adoption stick.
Read article : Why AI Training Doesn't StickAI Agents Are Coming for Knowledge Work: A Trainer's Field Notes
AI agents are moving from demos to deployments. What enterprise teams actually need to know about agentic workflows and what's still hype.
Read article : AI Agents Are Coming for Knowledge Work: A Trainer's Field NotesThe AI Maturity Trap: Why Most Companies Are Stuck at Stage 1
Most companies think they're behind on AI. They're right -- but the fix isn't more tools. The 4-stage maturity model and how to move past Stage 1.
Read article : The AI Maturity Trap: Why Most Companies Are Stuck at Stage 1The Skill Formation Problem: Is AI Making Us Worse at Our Jobs?
Research shows AI can negatively impact skill formation. Why that finding matters and what responsible corporate AI training looks like.
Read article : The Skill Formation Problem: Is AI Making Us Worse at Our Jobs?How PolyU's Finance Team Built an AI Workflow That Actually Works
Building sustainable AI workflows for a university finance office - starting with strategy, establishing the ACE framework, and watching tool users become process designers.
Read article : How PolyU's Finance Team Built an AI Workflow That Actually WorksTraining & teaching
Course design, classroom practice and the limits of a single workshop.
How to Choose an AI Train-the-Trainer Program in Hong Kong
Hong Kong is short of people who can deliver AI training properly. What a train-the-trainer program should hand you, the five questions to ask before you pay, and where I stand as someone who runs one.
Read article : How to Choose an AI Train-the-Trainer Program in Hong KongThree US Businesses, One Hong Kong Trainer
Between April and June 2026, I ran three remote AI coaching sessions for US businesses -- a realty firm, a skincare founder, and an equipment dealer. Here's what makes a single-session coaching call produce something worth the time.
Read article : Three US Businesses, One Hong Kong TrainerFive Things Every AI Trainer Should Prepare -- and the One That Saves You
After enough broken demos and quiet rooms, the preparation for any AI training engagement crystallized into five outputs. Here's what they are.
Read article : Five Things Every AI Trainer Should Prepare -- and the One That Saves YouWhat a 500-Person AI Webinar Can and Can't Do
I delivered a firmwide AI awareness webinar for a global advertising group in Hong Kong. About 500 people. Here's what a broadcast that size can realistically do, what it can't, and the three design choices that make it worth running.
Read article : What a 500-Person AI Webinar Can and Can't DoWhy My Best AI Workshops Have Thirteen People, Not Five Hundred
The industry counts reach. But the AI training sessions where workflow actually changes have thirteen people in them, every screen visible, every problem solved in real time.
Read article : Why My Best AI Workshops Have Thirteen People, Not Five HundredTwo Workshops, One Day: Parents in the Morning, Kids in the Afternoon
A parent workshop about AI fears at a Hong Kong school in the morning. A children's AI build in the afternoon. Same tools, same city, and the working answers I've started giving parents since.
Read article : Two Workshops, One Day: Parents in the Morning, Kids in the AfternoonPlus, Minus, Multiply, Divide: A Cheat Sheet for Teaching Copilot
I rewrote a finance Copilot workshop at 1am and reorganised it around four arithmetic moves. The framework outlasted the panic.
Read article : Plus, Minus, Multiply, Divide: A Cheat Sheet for Teaching CopilotThe Traffic Light Protocol: How I Teach AI Safety in 15 Minutes
A 3-color data classification system that makes AI safety intuitive. Used across banking, retail, and 500-person webinars.
Read article : The Traffic Light Protocol: How I Teach AI Safety in 15 MinutesWhy I Teach Copilot Instead of ChatGPT in Enterprise Workshops
Most AI trainers default to ChatGPT. Here's why I start with Microsoft Copilot for enterprise workshops and how it changes adoption.
Read article : Why I Teach Copilot Instead of ChatGPT in Enterprise Workshops
AI found the lowest quote. Why can’t we place the order yet?
A fictional quotation exercise shows how to check sources, missing information and decision boundaries.
Read article : AI found the lowest quote. Why can’t we place the order yet?Tools & workflows
Tools and working methods, grounded in what I use myself.
The most-used AI tools are not the ones people pay for most: what a16z's August 2026 lists show
Do people pay for the AI they use most? a16z's three Top 50 lists for August 2026, side by side: 29 of the 50 highest-spending consumer AI apps are not in the top 50 by mobile users or web visits, and only seven are on all three lists.
Read article : The most-used AI tools are not the ones people pay for most: what a16z's August 2026 lists showFrom a plain white page to a game lobby: how a 2001 game's look got me to open my ADHD to-do app
My ADHD to-do canvas used to look like an Excalidraw or Canva sketch: clear, and easy to ignore. My AI agent gave it the look of a 2001 game lobby, and the playful surface really helped me engage. The before and after, the method, and how to make your own skin.
Read article : From a plain white page to a game lobby: how a 2001 game's look got me to open my ADHD to-do appI asked my AI agent to find Christmas flights. The cheapest fare was not the answer.
My partner suggested Europe in December. My AI agent checked my calendar alongside the search, ran four agents in parallel on fares and airlines, and brought back receipts. The cheapest fare was not the answer. I decide.
Read article : I asked my AI agent to find Christmas flights. The cheapest fare was not the answer.I've Written Down 103 Rules for My AI. Three of Them Broke Last Week.
I published the guide to building a four-tier AI memory system. Then I spent three weeks grading my own. Three rules broke that were already written down -- and none of them failed because the system couldn't find the rule.
Read article : I've Written Down 103 Rules for My AI. Three of Them Broke Last Week.Why My AI Wrote Badly in Three Languages (It Wasn't the Model)
I asked my AI assistant why its writing is bad in English, Chinese and French. The answer wasn't the model -- it was the templates I fed it months ago, when the models were weaker. We deleted eight of them this week.
Read article : Why My AI Wrote Badly in Three Languages (It Wasn't the Model)The architect that never writes code
Companion take to the build log: the most expensive model in the stack wrote zero implementation code. Four bugs died at the spec stage, three workers passed the same five gates, and the blind pick landed on the cheapest. The pattern, not the model, is the asset.
Read article : The architect that never writes codeThree models, one branch, zero deployment
First live run of the architect pattern on my own website: Claude Fable 5 wrote the specs and verdicts, Grok 4.5 and GPT-5.5 did the typing in parallel. 75 files, +167/−668 — and nothing ships until I decide.
Read article : Three models, one branch, zero deploymentAI made me a manager, not a better maker
AI raised my output and quietly changed the job — from making the thing to deciding what is worth making. Five practitioners circling the same shift, and why taste is the part that did not get automated.
Read article : AI made me a manager, not a better makerThree Gemini Tools Collapsed Into One Prompt Today
Google shipped Gemini Omni at I/O 2026 — Veo, Nano Banana, and Genie folded into one multimodal prompt surface. Three of my daily Gemini calls collapse into one. Day one read, API still weeks out, the consolidation thesis lands.
Read article : Three Gemini Tools Collapsed Into One Prompt TodayHow to Prep for a Job Interview Using AI in 30 Minutes
A 30-minute AI loop for prepping a job interview — NotebookLM podcasts of your CV, then the company, then a recorded ChatGPT voice-mode mock interview.
Read article : How to Prep for a Job Interview Using AI in 30 MinutesWhy Hong Kong's AI Disadvantage Is Actually an Advantage
I trained 120 CEOs in LA who couldn't copy and paste. Hong Kong's tech literacy is higher than people think. The real gap isn't VPNs — it's information lag and the Cantonese blind spot.
Read article : Why Hong Kong's AI Disadvantage Is Actually an AdvantageMy Blog Publishes Itself. It Died Silent for 3 Weeks.
I built an autonomous AI blog pipeline. It ran for months, published 47 posts, then silently stopped for three weeks. Exit code 0. No alerts. Here is what went wrong and the three principles I now follow.
Read article : My Blog Publishes Itself. It Died Silent for 3 Weeks.I Built an AI Chatbot in 2 Weeks. Here's What Broke on Turn 3.
The gap between 'build an AI chatbot' tutorials and a working production chatbot is measured in crashes, security holes, and spec rewrites.
Read article : I Built an AI Chatbot in 2 Weeks. Here's What Broke on Turn 3.Claude Code Mastery: The Full Blueprint — My Actual System
I open-sourced my entire Claude Code architecture. Every file, every connection, every lesson learned. 9 agents, 40 skills, 7 cron jobs, 4-tier memory — built because I have ADHD and needed external structure.
Read article : Claude Code Mastery: The Full Blueprint — My Actual SystemAI Change Management: What Actually Works (A Field Report From 70+ Organizations)
Most companies approach AI adoption like a software rollout. After training 10,000+ professionals across 70+ organizations, here is what actually drives lasting behavior change.
Read article : AI Change Management: What Actually Works (A Field Report From 70+ Organizations)Claude Code Mastery: Agent Teams & Automation — Your AI Workforce
I have 9 AI agents running 7 cron jobs. My blog publishes itself every Monday. Here is the full agent team architecture — delegation rules, parallel execution, and cost-aware model selection.
Read article : Claude Code Mastery: Agent Teams & Automation — Your AI WorkforceClaude Code Mastery: Hooks & Guardrails — Making AI Safe
My AI agent nearly sent an email to the wrong person. That one near-miss changed everything. Here is the hooks system I built — PreToolUse validation, prompt injection scanning, and a policy engine.
Read article : Claude Code Mastery: Hooks & Guardrails — Making AI SafeClaude Code Mastery: Skills & Memory — Teaching AI to Remember
AI forgets everything between sessions. I solved this with a 4-tier memory system and 40 active skills that load in 100 tokens each. Here is exactly how persistent knowledge works in Claude Code.
Read article : Claude Code Mastery: Skills & Memory — Teaching AI to RememberClaude Code Mastery: From Zero to Your First Agent
Three months ago I opened Claude Code and thought it was just another terminal. Today it runs my entire business. Part 1 of a series on how I actually use it — installation, mental model, CLAUDE.md, and your first real task.
Read article : Claude Code Mastery: From Zero to Your First AgentThe Agent Adoption Ladder: What Most People Get Wrong About AI Agents
I taught my non-technical co-founder about AI agents. What came out is a four-tier framework for matching the right agent tool to how you actually work — from chatbots to multi-agent orchestration.
Read article : The Agent Adoption Ladder: What Most People Get Wrong About AI AgentsWhat Nobody Tells You About Building Agentic AI
I built a multi-agent AI system from scratch -- four specialists, a team router, persistent memory, and streaming UX. Here are five things the framework docs don't cover.
Read article : What Nobody Tells You About Building Agentic AII Rebuilt My Entire SEO in One Day. Here's What Actually Moved the Needle.
How I used Claude Code to rebuild my entire SEO and GEO strategy in one session -- structured data, internal linking, hreflang, and pillar pages across 30+ blog posts in 7 hours.
Read article : I Rebuilt My Entire SEO in One Day. Here's What Actually Moved the Needle.5 AI Prompts That Actually Help With Procrastination (From Someone Who Uses Them Daily)
Five viral AI prompts for procrastination -- why each one works, how to get more from them, and what changes when you have ADHD and use them every day.
Read article : 5 AI Prompts That Actually Help With Procrastination (From Someone Who Uses Them Daily)What the 2026 Budget's HK$50 Million AI Training Push Actually Means for Your Company
Hong Kong earmarked HK$50 million for AI training in the 2026-27 Budget. Here's what it actually means for corporate training, what's missing, and what to do now.
Read article : What the 2026 Budget's HK$50 Million AI Training Push Actually Means for Your CompanyWhat 1,400 People Taught Me About the Gap Between AI Demos and AI Adoption
I gave a live talk on AI agents to 1,400 registrants. The audience reaction confirmed what I see in every training room — people want proof, not promises.
Read article : What 1,400 People Taught Me About the Gap Between AI Demos and AI AdoptionThree AI Gaps in an E-Commerce Team Already at 70-80% Adoption
A Shenzhen visit exposed a pattern I see everywhere: teams using AI at 50% efficiency because nobody taught them model selection.
Read article : Three AI Gaps in an E-Commerce Team Already at 70-80% AdoptionI Built an RTK-Style Strategy Game on Top of My AI Workflow
Every tool call in my AI assistant logs to SQLite. So I turned it into a Romance of the Three Kingdoms-style strategy game -- with named skills, fractional XP, and a six-agent squad.
Read article : I Built an RTK-Style Strategy Game on Top of My AI WorkflowI Built a Voice Mode for Claude Code Because I Can't Read
Long terminal output is attention-hostile for ADHD developers. So I built a plugin that speaks Claude Code responses aloud -- zero deps, one install command.
Read article : I Built a Voice Mode for Claude Code Because I Can't ReadHow an Art Prompt Turned Into a 78-Bug Audit of My AI Assistant
I asked my AI to draw what's in its mind. The interesting part wasn't the drawing -- it was the working session that followed, where we found 78 logged misfires and shipped 14 fixes.
Read article : How an Art Prompt Turned Into a 78-Bug Audit of My AI AssistantThe AI Training Gap: 70% of Managers Want It, 14% of Workers Get It
70% of managers say an AI-trained workforce is critical, but only 14% of workers have been offered training. Here's what's going wrong.
Read article : The AI Training Gap: 70% of Managers Want It, 14% of Workers Get ItWhy Your IT Department Is Your Biggest AI Adoption Bottleneck
Enterprise AI adoption fails when IT blocks everything by default. How to turn IT from gatekeeper to enabler using a 3-tier security framework.
Read article : Why Your IT Department Is Your Biggest AI Adoption BottleneckStop Telling AI to 'Be Creative': What 35 Prompting Strategies Actually Reveal
Wharton researchers tested 35 prompting strategies and found that saying 'be creative' is the least effective. Here's what actually works.
Read article : Stop Telling AI to 'Be Creative': What 35 Prompting Strategies Actually RevealWhat Happens When a Toy Company Takes AI Seriously
Two sessions, 12 executives, and a gardening demo that secured buy-in. Inside an international toy company's AI training.
Read article : What Happens When a Toy Company Takes AI Seriously84% of People Have Never Used AI. That's Your Competitive Advantage.
84% of people have never used AI and only 0.3% pay for it. The window of competitive advantage is wide open.
Read article : 84% of People Have Never Used AI. That's Your Competitive Advantage.The AI Tool Stack I Use in Corporate Training
The complete AI tool stack an AI trainer uses across banking, retail, education, and engineering workshops -- from Copilot to NotebookLM.
Read article : The AI Tool Stack I Use in Corporate Training3 AI Tools That Saved My Focus (And Can Save Yours)
Three specific AI tools that solve real attention problems: information triage, meeting recovery, and structured thinking. Based on my second appearance on Club 80.
Read article : 3 AI Tools That Saved My Focus (And Can Save Yours)Why Free AI Isn't Enough: The Case for Paying for AI Tools
76% of viewers in a live poll said they don't pay for AI tools. Most of the 24% who do aren't using the features they paid for. The actual percentage using AI at full capability is vanishingly small.
Read article : Why Free AI Isn't Enough: The Case for Paying for AI ToolsFour Prompt Patterns from a 2.5-Hour Community Workshop
When Dorich asked for an AI session that wouldn't be forgotten in a week, we built it around four specific prompts their audience could use the next morning.
Read article : Four Prompt Patterns from a 2.5-Hour Community Workshop