Map the Task, Keep the Judgement, Then Ask for the Time
7 min read
The most useful question is not which course to take next, but which part of a job you already do would you be willing to hand to an assistant and still stand behind. Work backwards from there: pick one recurring task, split it into stages, mark the ones where you could check an output against a source, and keep the rest. A course only comes into the picture once you know which stage you want to get better at.
That question matters now because official support has moved towards applying AI to real tasks. CNA reported on 16 September 2026 that SkillsFuture Singapore's successor agency, SWDA, had rolled out more than 200 AI-related training courses with free six-month subscriptions to selected premium tools for eligible Singapore citizens. Acting Minister for Manpower Jasmin Lau framed the courses and tool access as a way to help people learn, practise and try new skills, while cautioning that nobody becomes an expert after a single course or workshop. A stated aim tells you where support is pointed; it does not choose a task for you. [1]
Start with the work, not the tool
Open a note and list what you actually produced in a real working week. Not your job title or your software, but the reports, briefing notes, first drafts, reconciliations and meeting write-ups, plus the decisions you made and the people you kept informed. Doing this from your own calendar rather than a product page tells you what to try first, because it starts from work that already has a quality standard attached to it. Then choose one item that recurs often enough for you to notice the pattern.
That starting point matches where ministers say change is landing. CNA reported in September 2026 that Minister Ong Ye Kung assessed AI as generally replacing certain tasks rather than entire jobs, and said it is used as a tool that can result in some job redesign. In the same report, a healthcare worker described using AI for record-taking during consultations and as an extra pair of eyes on diagnostic scans, while an interior designer used it for renderings but still coordinated contractors, applied for permits and liaised with clients. If change arrives task by task, a task inventory is where you see it coming. [2]
Mark the stages, not the whole task
Now split that one task into stages, because most real jobs are hybrids. A two-page summary, for instance, has a gathering stage, a structuring stage and a stage where you write the recommendation. Mark the stages where the hard part is assembling material, producing a first pass or comparing options against criteria you have already set. Those are worth practising on, because you can read the output back against the source. [3]
Beside them, mark the stages that stay yours: anything turning on context only you hold, judgement you would have to defend, relationships you maintain, or accountability you personally carry. A CNA commentary made the related point that as AI takes on information gathering, drafting and initial analysis, human work can shift towards interpreting information, making decisions, solving problems, exercising judgement and building relationships, and that the employee also has to decide whether to trust a recommendation and when to probe further. Recording both kinds of stage together is what keeps control visible inside a mixed task. [3]
Adapt the marking to your professional duties, regulatory responsibilities, employer policies and the sensitivity of the information involved. In a clinic, a school, a finance function or a care setting, some stages may be closed to you for good reason. A practical test for any stage: if you cannot describe how you would catch an error in it, you are not yet able to supervise it.
Settle review, ownership and workload first
Before running a first draft, agree four things separately with your manager. Who reviews the output, and at what point. Who signs it off and answers for it if it is wrong. What quality the finished work has to reach. And how much of your existing workload the trial displaces, rather than adds to. Keeping these distinct is what stops a trial quietly becoming unpaid extra work, because the workload question is the one that usually goes unasked. [3]
These are practical questions, and organisations face them too. The same commentary said organisations need to answer which tasks AI should perform and which remain human-led, how responsibilities should change, and how performance should be measured when workers use AI tools. It cautioned that if AI is simply added onto existing jobs, employees are expected to keep meeting current targets while also learning new tools, and it starts to feel like another demand rather than a tool that helps. [3]
To picture it: say you spend about forty minutes a week turning raw site notes into a two-page summary for circulation. Your plan is to let AI produce the first structure, check every figure and quote against the notes yourself, and rewrite the two paragraphs carrying your recommendation. The finished summary stays the quality measure; separately, ask how much time the checking will now take and whether that comes out of the rest of your week.
Turn the conversation into an arrangement
Take four requests into the same meeting: permission to experiment, with a clear view of what you will not send to an external service; time, in a fixed slot in working hours rather than the leftovers at nine at night; someone to learn alongside; and a manager who can coach rather than only supervise. The CNA commentary argued that meaningful support has to show up in how work is designed and managed, not just in courses and workshops, and that employers owe workers opportunities to apply skills, not only to acquire them. [3]
Offer something in return by bringing your marked-up task and asking to help decide what gets redesigned. The same commentary said managers who explain the purpose of a change, encourage experimentation, acknowledge uncertainty and involve employees in redesigning work are more likely to build trust than those who simply instruct staff to use more AI, and noted that many managers are navigating these questions for the first time themselves. Knowing that helps when a first conversation stalls. [3]
If your organisation wants to look beyond one person's one task, official signposting exists. SWDA's job redesign page lists the Workforce Development Grant for Job Redesign, described as providing enhanced support for workforce transformation through workforce consultancy, capability building and HR tech solutions, and a Job Redesign Centre of Excellence offering guided sectoral playbooks, HR resources on workforce planning and capability development workshops. SWDA's workforce and skills pages list TalentTrack+, described as digital tools to identify skill gaps, align training plans and map career pathways. These are optional support for organisations doing redesign work, not a step in your own exercise. [4] [5]
Once the map is done, you have something more useful than a shopping list. You know which stage you want to get better at, and you know what you will keep. If the gap is a skill — writing, data handling, domain knowledge — that is the point at which a course, a workshop or free tool access becomes a sensible answer rather than a first move. If the gap is permission or protected time, more training will not fix it.
Read next
Sources
- Singapore launches over 200 training courses with free AI subscriptions, new skills centre – CNA
- AI replacing certain tasks, but not entire jobs: Ong Ye Kung – CNA
- Commentary: Workers shouldn't have to figure out AI on their own – CNA
- Job Redesign
- Workforce & Skills | SWDA
BUTLER Magazine Editorial · AI-assisted research and writing, reviewed by our automated editorial team. Sources checked 2026-10-03. Featured image: AI-generated editorial illustration.
