The brief
After the break, the morning's theory became a build. Each group: pick a course you could actually run — he suggested making nasi lemak or safe driving for new drivers, "I always have cooking nasi lemak" — identify the audience, run a quick needs analysis, list the competencies, then design the course on the real templates: the course design table, the course content table with theory and practical hours, and a schedule. The constraints were HRD Corp's own: adult audience, maximum 25 pax for technical courses or 50 for non-technical, duration of four or seven hours, face-to-face or online. The templates arrived in the WhatsApp group, and the room split three ways: one group went outdoors, one went gamification, and our group went where it was always going to go — AI. Ninety minutes later we had a course I would genuinely sell.
Our submission — AI Empowerment
We presented "AI Empowerment — Building an AI-Ready Workforce" (working title on the design table: Hands-On AI Jumpstart): a two-full-day corporate programme for working professionals, recommended twenty pax. Description as a proper paragraph — practical and hands-on, equipping employees to use generative AI effectively in the workplace. Competencies: AI literacy, prompt writing, digital productivity, responsible and confident use of AI. Training objective: apply generative AI tools to complete workplace tasks accurately, productively and responsibly. Six learning outcomes, from explaining fundamentals through constructing effective prompts to designing an AI-assisted workflow for a real workplace scenario. Methods: lecturette, live AI demonstration, hands-on labs in pairs, buzz groups, workplace simulations. Assessment: practical exercises against an observation checklist — my pitch line was that by the end, "you will be able to use AI to generate what you want within half an hour." Eight modules from AI fundamentals to responsible AI and a personal action plan, split almost exactly 50/50 between theory and practical hours. The live demo of course-building had one honest bug: our schedule slide showed one day of eight hours for a course we had just declared was two full days. "I was wrong — later it will change." Change on the go, as one does. The room also got its joke in early about our group's methods — that on Friday I would just stand in the corner while "the AI ran the whole day." Fair. Duly noted for my session plan.
The module table
The engine room of the submission was the content table: eight modules, from AI fundamentals and workplace impact through prompting, productivity, content creation, analysis, problem solving and automation, closing on responsible AI and a personal action plan — each with a one-line description and its theory and practical hours split out, exactly the format from the morning's outline module. Fifteen hours total across the two days: seven of theory, eight of practical. Writing it, I felt the top-down sequence working the way Martin described his own: the six programme outcomes gave us the module list almost mechanically — each outcome claimed one or two modules, and any module that couldn't point at an outcome didn't survive the draft. The table is really the programme's outcomes wearing a timetable.
Martin's red pen — prompting AI for outcomes
The feedback began kindly — description in paragraph form, "very good"; overall structure solid — then went where it needed to go. Some sections too wordy: "short and sweet." And six learning outcomes drew the sharper question: can a participant actually tell two of them apart, or are several the same outcome in different clothes? "You need to be mindful if two learning outcomes — can people differentiate what is the objective from these two outcomes, or is it actually more or less the same?" He assumed, correctly, that AI had helped generate them — "your eye is already clever" — and turned it into the morning's most practical lesson. Early AI-generated outcomes, he said, were trivially easy to spot: they all opened with the same four verbs — learn, appreciate, know, understand. The unmeasurable quartet, the exact words Day 2 taught us to ban. His fix is a prompting protocol: tell the AI explicitly to follow Bloom's taxonomy; constrain it to the level you are actually teaching — "if you want only level one and two, AI can generate only learning outcomes for level one and two"; demand the verb–object–qualifier structure from Day 2's outcome-writing module; then "ask the AI to generate ten possible learning outcomes. I take the ones which are next to it — I never use from AI directly. I just get the idea. I think through the learning outcomes and then I restructure it a bit." Generate wide, choose, rewrite by hand. He runs this on all his programmes, down to OKR workshops. The last cut was our 50/50 theory–practical split. Acceptable to HRD Corp, "but more theory than practical is not good" for a skills-based fund — 60/40 is better, and 70% practical is the competency-standard benchmark. For a course whose entire premise is hands-on, we had quietly built a lecture-heavy timetable. The red pen sees what the builder can't.
The other groups
The outdoor group pitched a four-hour backpacking course at Taman Teknologi — competencies including communication, teamwork, equipment selection and, wonderfully, "able to have fun and sense of humour." Their learning outcome earned real praise: at the end of the session, participants will be able to pack, organise and select efficient camping equipment within 30 minutes — a genuine ABCD outcome with a condition and a measurable degree, the full Day 2 formula in the wild. Then the scalpel: that sentence hides three verbs, which means it is secretly three learning outcomes — pack, organise and select are different skills wearing one outfit. And a clarification that un-stuck our own thinking about Friday: individual assessment does not mean one person at a time. Give four participants four equipment sets and assess all four simultaneously — individually scored, not individually queued. Their schedule got the same note ours would have: five-minute breaks don't belong in a submitted schedule. The gamification group's "Game On" — a seven-hour course teaching trainers to design original training games through prototyping and play-testing — came out the cleanest: more practical than theory, sensible outcomes, a play-test as the assessment. Two touches of the pen: "know what game to use in what scenario" should be identify — know is on the banned list even inside a competency statement — and the hour-by-hour schedule should relax into blocks: "just make it ten-to-twelve. No need one hour, one hour." Watching three groups get three different corrections was its own masterclass: the frameworks are shared, but every design fails in its own way.
One good learning outcome
The morning closed with the recap slide — the seven steps from performance standard to assessment planning — and a final calibration for Friday that Martin clearly repeats every cohort and clearly gets ignored every cohort. A comprehensive training programme, he said, ideally carries at least one learning outcome per domain: one cognitive, one psychomotor, one affective — because a room full of different learners needs different kinds of change. That is the professional standard for a real course. Friday is not a real course. Friday is fifteen minutes, and fifteen minutes gets exactly one learning outcome: "Write one — but one very good learning outcome. I say this every time on Wednesday, and on Friday I still get people with three learning outcomes. Three learning outcomes in fifteen minutes is not measurable. I give you not full marks. I'll be very straight on that." One well-made outcome, fully aligned and fully achieved, beats three ambitious ones that all die unassessed. It is the horse puzzle again at smaller scale: fewer pieces, arranged perfectly.
Case Study
the exercise was the audit
The strange loop of today: the course our group designed is the course I actually teach. So tonight the classroom exercise became a production audit of the real DJC AI course, applying Martin's red pen line by line. First, the LAKU sweep. I ran the banned-verb check across my actual course page and slide deck: two module descriptions promise that participants will "understand" things, and one opens with "learn." Rewritten tonight with the prompting protocol exactly as prescribed — Bloom's taxonomy named in the prompt, constrained to apply-and-below for the fundamentals module and analyse-and-up for the workflow module, verb–object–qualifier structure demanded, ten candidates generated, three chosen, all three rewritten in my own words. The AI drafts; the trainer decides. Second, the ratio: my one-day format runs closer to 60/40 theory-heavy — the inverse of the benchmark. Two lecture segments become labs: the prompting principles talk becomes a prompt-rewriting exercise on the participants' own real emails, and the tools overview becomes a fifteen-minute guided scavenger hunt. Third, the document split I had never made: my current proposal PDF is a session plan wearing a brochure's clothes — it lists activities minute by minute, which after today reads less like transparency and more like handing out my choreography. The public version shrinks to module names and time blocks; the run sheet goes private. And one structural fix straight from the backpacking group's feedback: my flagship outcome — "participants will be able to prompt, evaluate and automate with AI" — is three verbs in a trench coat. It is now three outcomes, each with its own activity and its own assessment, which instantly exposed that automate never had an assessment at all. Constructive alignment found the gap in under a minute. That is the fastest consulting engagement I have ever received, and I was both the consultant and the client.
Reflections
Ninety minutes to design a sellable course sounds impossible until you notice what was already in the room: three days of frameworks queuing up to be used. The TNA vocabulary told us who the course was for, Bloom's verbs wrote the outcomes, the methods catalogue filled the schedule, and constructive alignment kept the three corners honest. The templates didn't teach us anything new — they revealed how much the previous days had already installed. The deeper lesson was watching a professional use AI in exactly the posture I teach but don't always practise: wide generation, narrow selection, human rewrite, named constraints. "I never use from AI directly" is not AI-scepticism — it is the trainer keeping the one job that cannot be delegated: judgement. Tomorrow afternoon is session planning, tonight my course got redesigned by a classroom exercise, and Friday everything above compresses into fifteen minutes with one verb. Arrange the cards well.