Singapore has set an ambitious target: 100,000 “AI-bilingual” workers by 2029. The goal signals a broader reckoning with how AI is reshaping the professional workforce — not merely as a productivity tool, but as a capability that demands a new kind of literacy. Yet as training programmes multiply and certification frameworks take shape, a harder question is emerging: what does AI bilingualism actually require in practice?
“AI bilingualism means having enough domain expertise and AI fluency to actually direct, evaluate, and push back on what AI gives you,” says Felicia Tan, Director of Tribe Academy, in an email interview with e27.
The distinction matters. Faster, more polished output is already well within reach for most professionals. The ability to spot where that output is wrong — or quietly dangerous — is proving far more elusive.
Tribe Academy offers expert-led training in areas including AI and blockchain to further bridge Singapore’s talent gap. In our conversation, Tan reveals the blockers that many corporations in Singapore face in embracing AI–and what to do about it.
The following is an edited excerpt of the conversation.
Also Read: 5 Seoul startups made their Southeast Asia debut at Echelon Singapore 2026 under the SBA pavilion
MOM’s latest survey shows 70 per cent of Singapore companies still have not adopted AI for work, a striking number given how much policy attention has gone into this space. In your experience working with corporate clients, what’s the real blocker?
If you spend a lot of time in tech circles, it can feel like everyone is already using AI. But outside that bubble, many organisations are still at the stage of observing, experimenting cautiously, or waiting to see clearer proof of value before changing how work gets done.
Everett Rogers gave us the Diffusion of Innovations curve decades ago, and it remains one of the most useful lenses for moments like this. The theory has long shown that every major shift moves through stages. Innovators, early adopters, early majority, late majority, then laggards, each arriving on their own schedule, for their own reasons. Right now, AI still sits heavily between the early adopterand early majority phase for many Singapore companies.
From the perspective of early adopters, it can feel like progress is slow. But we also need to recognise the scale of behavioural change being asked of the workforce. The oldest members of our working population in Singapore today entered their careers roughly 30 years ago, in the mid-1990s … Entire careers were built around ways of working that rewarded precision, hierarchy, and predictability.
AI changes not just the tools people use, but the nature of how work gets done. That transition naturally takes time, especially at workforce scale. Policies and national initiatives help create momentum, but cultural and operational change inside organisations has always moved slower than headlines.
One main blocker we are seeing with AI adoption is that it is still highly siloed and deeply individual. We see individuals attending our programmes who bring these skills back, but only to their personal chat windows. Someone on the team discovers a prompt that saves them two hours a week, and they quietly use it, and nobody else knows. Someone in HR uses a new AI tool for meeting summaries but the knowledge stays private. There is no institutional memory layer or shared playbook that captures what’s working.
Also Read: The great rotation: How AI stocks are stealing billions from crypto
So you get a patchwork where a few power users produce impressive outputs, while everyone else is doing things roughly the way they always have. The gain will live and die with the individual. For organisation-wide impact, a deliberate redesign of workflows, KPIs, or operating models will need to follow either through top-down directives or a conscientious effort by the entire staff.
The next blocker is arguably the most honest one, i.e. if it isn’t broken, why fix it? Not everyone is a productivity advocate and lies awake worrying about workflow inefficiency. Sure, some firms are redesigning roles and creating new AI-related positions. But AI is not visibly taking anyone’s job tomorrow. Then reasonably, as human beings, we tend to stay with what works, that is, no change is needed.
The third blocker is structural among those who have attempted to look into AI implementation. The most commonly cited constraints are high implementation costs and lack of in-house expertise. The tools exist. The willingness, in many cases, exists too. But the bridge between “I’ve heard of AI” and “We’ve redesigned our workflow around it” is still too long and too expensive for most SMEs to cross without support.
There’s a tendency among early movers to look at policy timelines and grow impatient. But policy takes time to land. The government has announced a new Tripartite Jobs Council to support employers and employees in AI adoption, alongside access to free premium AI tools for Singaporeans taking selected AI courses.
The reality is that a Budget announcement in February does not transform workforce behaviour by April. There will always be a lag between national intent and organisational habit change. The real test is what companies do in that gap. Grants and national initiatives can reduce the risk of taking the first step, but they cannot redesign workflows on behalf of every employer.
Also Read: Agentic commerce’s dirty secret: the data powering AI purchases is often wrong
There’s been a lot of hiring around “prompt engineers” over the past two years. Is that the right unit of skill to be built for? What’s the capability that actually drives business value that most job descriptions and course catalogues are still missing?
Good prompt engineering is underrated. It was the first core skill that emerged when Generative AI became accessible to the general public, and it remains foundational.
The fact that leading AI companies are still publishing prompt engineering guides and running 101 courses around it tells us something. The practical case for why it matters more is important as models get more powerful. The newer reasoning models consume significantly more tokens, especially when you are building complex workflows or automating multi-step processes. If you do not know how to construct a tight, well-structured prompt, you will burn through credits at an alarmingly fast rate.
Extrapolating this across a team running dozens of automated workflows, and it becomes economically untenable. It is a boring skill compared to flashy AI apps and dashboards, but knowing how to communicate with AI systems precisely will save companies enormous amounts of money and frustration over time.
Prompt engineering remains a useful skill to develop, but its real value is as a foundation for broader AI capability, not as the end goal. Prompt engineering gets one to a useful first draft. What you do with that draft … is one that most job descriptions and course catalogues are still fumbling to articulate, but what is going to derive the most business value.
So, if I were advising an enterprise on what capability to actually build for, it would be this: Workers who have developed strong prompt discipline as a baseline habit, and who pair it with critical thinking to know when the model is leading them somewhere wrong. The future is probably less “prompt engineer” and more “AI-native operator”.
Also Read: Why Cyber Risk Ownership Is Southeast Asia’s Biggest Leadership Blind Spot
If you could redesign one thing about how Singapore is approaching workforce AI upskilling right now, what would it be?
If I could redesign one thing about how Singapore is approaching workforce AI upskilling right now, I would shift the focus from primarily funding structured training to creating a much stronger bridge between training and rapid on-the-job application.
Many companies still see upskilling as “time away from real work”. To close this gap, we need structured training to remain the foundation in building mental models and tool confidence, but we must also ensure it then quickly flows into real application. Specifically, companies should set aside dedicated hours each month for employees to test AI on actual tasks, just like how R&D time is ringfenced in tech companies and “timetabled time” is set aside for teachers to dedicate time to innovation and professional development.
Policy can reinforce this by tying enhanced grants to organisations that implement and report on these pilots, with even stronger support when they become sustained initiatives rather than one-off efforts. Once organisations have a core group of upskilled champion users, they should guide the rest of the team to start small with specific tasks, such as shortlisting documents, summarising meeting notes, or automating a two-part workflow.
The goal is to learn by doing something real and low-stakes. Have employees treat early failures as cheap tuition. Just like in the early days of the internet, nobody expected the first company website to generate revenue immediately.
This approach aligns with one of Singapore’s strongest policy philosophies: reducing the downside ofexperimentation. Our grants, co-funding, and training subsidies were never designed to guarantee perfect outcomes, they exist to make the first step less risky and encourage early action.
By redesigning the system this way, we can turn awareness and training into genuine productivity gains and keep Singapore’s workforce truly competitive.
—
Image Credit: Cash Macanaya on Unsplash
The post Tribe Academy’s Felicia Tan: Why good prompt engineering and critical thinking are keys to AI bilingualism appeared first on e27.