Where Capability Lives

5 May 2026 - Skills and Workforce Development Agency Bill, Second Reading Debate

Speaker,

We are asked today to approve the merger of two agencies. I would use this occasion to ask whether the system they inherit is the right one.

We often speak of skills upgrading and credentials. The real question is simpler: are we building real capabilities for our workers?

Here’s where I think the Government has done well.

Workforce Singapore’s Place-and-Train Career Conversion Programmes are — by the Government’s 2024 evaluation — a success. Close to eighteen thousand mid-career Singaporeans were placed between 2017 and 2022. Wages rose by up to six and a half per cent by year four. Employment retention improved by four percentage points. And the biggest gains went to non-PMET and mature workers — the groups the credit-based system has struggled to reach.

I also would like to acknowledge the 38 Skills Frameworks - co-built with industry, unions, and professional bodies. They are a useful compilation of the skills and competencies expected for each role in these sectors. But as those who’ve been hiring managers know: what’s in the frameworks sets necessary but insufficient bars. You still need interviews and competency tests to actually hire well.

This success of this place-and-train, employer-anchored programme contrasts with other parts of the system which are not doing so well, such as the credit-based, individual-based SkillsFuture programme.

(The Principal-Agent Problem with SkillsFuture)

The credit-based side of the system has struggled on both integrity and take-up. In 2017, a syndicate fraud extracted nearly $40 million through fake enrolments. The Auditor-General later found millions more in overpayments and uncollected levies. On demand: the SkillsFuture Level-Up Programme — the flagship of mid-career upskilling, offering a $4,000 top-up — had reached only 28,000 out of 1.2 million eligible citizens by end-2024. That is two per cent.

I don’t think this is an execution failure, but a flaw in the policy idea itself.

The SkillsFuture system of the last decade is an example of the individual-credit system: the government funds individuals to purchase courses from a permissive provider market. Suppliers capture the funds; fraud is possible; links with entry-level capability are weak. The largest meta-analysis of active labour-market programmes — Card, Kluve, and Weber, 2018, with 207 estimates across 200-plus studies — found classroom training shows weak or negative short-run effects, with gains emerging only at the two- to three-year mark, and concentrated on the long-term unemployed.

The populations best served by these programmes are the displaced and long-term jobless — not mid-career workers who need to shore up against job precarity or climb the wage ladder.

In France, they tried this model. In 2015, under President Hollande, they launched the Compte Personnel de Formation — a Personal Training Account. Training hours attached to the person, not the job. The 2018 Macron reform converted those hours to euros — €500 a year, capped at €5,000, with higher caps for the lower-qualified. Then in 2019 they built an app — Mon Compte Formation — a consumer-grade marketplace where any worker could browse, book, and pay for any accredited course in a few taps. Employers and unions were cut out entirely. Twenty-one million accounts were activated. More than two million training purchases a year. It was one of the boldest individual-credit experiments of any major economy.

The result is instructive. TRACFIN — France’s financial intelligence unit — recorded a five-fold rise in suspected fraud in a single year, from €8 million in 2020 to €43 million in 2021. Cold-callers and fictitious providers followed the money. France spent the next four years walking it back — banning cold-calling, imposing co-payments that have since been raised twice, capping the most-abused course categories. France had the digital infrastructure. It had the regulators. It had the financial intelligence unit. What it did not have was any party other than the state itself bearing consequences for training quality. The principal-agent problem is not solved by transparency, digitisation, or individual choice.

The chain of transmission of consequences for an individual credit system, from ministry to agency to provider to learner to employer, is such that nobody bears the consequences of low-quality training. It is a principal-agent problem. No one is a principal. The Career Conversion Programmes work because an employer is the principal before training begins. The individual-credit channel, by design, lacks that principal, and no amount of enforcement can substitute for it.

It seems SkillsFuture has unfolded in a similar way.

Can we redesign the system so that employers — not individuals purchasing courses — become the accountable parties? Not individual choice, which is what we have, but structured employer accountability: sectoral associations and firms bearing reputational and financial consequences for training quality. This would sharpen the Skills Frameworks too - employers with skin in the game will push for a fuller accounting of what a job actually requires — not just entry criteria, but the tacit knowledge needed to stay and grow.

When funds are provisioned individually, the government defaults to input-side accountability: ensuring providers are accredited, claims meet thresholds, and participants receive a certification. Whether the credential improves a worker’s employability is not measurable at the point of purchase — only in labour outcomes, years later. The system optimises for what it can count, not what it should produce.

I think it is desirable to have sectoral associations and firms lead the upskilling journey directly, to solve the principal-agent problem, and consequently, invest in deep, long-duration training.

(Policy Space to solve the Principal Agent Problem)

How do other economies solve the principal-agent problem?

Germany’s apprenticeship system operates within one of Europe’s most mobile credentialed labour markets, their IHK [Industrie- und Handelskammer] or Chambers of Industry and Commerce certification is portable. The goal is workers who are chosen repeatedly at a premium, not workers who cannot leave. German apprentices earn ~15% more than the untrained two decades after entry, with lower unemployment risk and faster re-employment.

Economic skill-formation literature identifies three conditions for firms to invest in deep, long-duration training:

First, wage compression — sectoral bargaining that narrows pay differentials within a sector so rivals cannot simply poach trained workers at a premium.

Second, portable credentials — standardised, externally certified occupational profiles, so a worker’s skill is legible to the whole labour market and the worker is willing to invest their own time.

Third, industry co-governance of curriculum — so that what is taught tracks what industry actually needs.

Singapore has the second of these, roughly. SkillsFuture certifications approximate portable credentials. Industry Transformation Maps and Trade Associations do consultative work that touches co-governance, but it is advisory, not binding — no sectoral employer body exercises a sign-off or veto on SkillsFuture curriculum, and no named individual’s reputation moves with and is staked on the employment outcomes.

As I understand it, NTUC mainly bargains at the enterprise level rather than at the collective sector level.

So I think we have neither true industry co-governance of curriculums, nor do we have sectoral bargaining to achieve wage compression.

Some will say Singapore is too small for German-style sectoral institutions. Switzerland — population nine million — sustains deep sectoral bargaining and apprenticeship in pharma, watchmaking, and machine tools. Size is not the binding constraint. Institutional architecture is.

I have two proposals.

First, sectoral bargaining. NTUC’s current architecture is attuned to individual employer engagement. That is not the right vehicle for deep tacit-knowledge sectors. These sectors need binding sectoral wage scales — not just floors, but scales that compress the wage distribution.

Right now, a firm that invests in deep training risks losing that worker to a competitor the day after certification. Wage compression fixes this — it narrows pay gaps so poaching is not worth it. The return on training stays with the firm that trained. That is why German firms fund three-year apprenticeships at their own expense.

The Progressive Wage Model lifts wages for lower-wage workers, and it does so well. But it sets floors, not scales. It says to the worker: train, and you move up. What we need is a structure that says to the employer: train deeply, because you will keep the return.

Not every sector is suitable. Sectoral bargaining fits sectors with deep tacit knowledge — semiconductor back-end, aerospace MRO, petrochemicals, marine and offshore — perhaps not faster-moving industries like AI.

Second, co-governance with consequences. Named industry entities and individuals should have sign-off and veto on SkillsFuture curricula — not as advisors, but as accountable parties whose reputations move with employment outcomes. We must also lower the accreditation barriers that keep the most credible practitioners out. The Advanced Certificate in Learning and Performance requires 71 hours of coursework — screening out the domain experts we need and screening in those with time to spare.

(What would be suitable for other faster-moving industries - where sectoral bargaining cannot be done?)

For faster-moving industries — AI, software, frontier hardware — the fixes above may not work. Skills change too fast; firms are too few and too mobile. What works in these sectors is something different: knowledge spreading through people working near each other. That is how Shenzhen and Silicon Valley were built — rapid knowledge diffusion through firms, suppliers, and people solving problems together, carrying what they learned into the next job.

We cannot build a Shenzhen-scale agglomeration on 750 square kilometres. But we can do two things.

(Industrial commons here)

One, engineer knowledge diffusion through people, not papers.

To produce an industrial commons here, first, we will need to revive the engineering society ethos. If we are serious about capability over credentials, the most basic question is: do our workers and engineers know how things work? We need to know how things work — at the level of teardown and re-assembly — because that is the foundation of any serious industrial capability. The model is Munro & Associates in Michigan, which has spent close to four decades disassembling competitor vehicles down to the last fastener and producing the detailed cost, design, and process reports that the global automotive industry treats as authoritative. Japan’s National Institutes of Technology teach fifteen-year-olds to disassemble electromechanical systems in formal “Reverse Engineering” courses. The US military reverse-engineered the Iranian Shahed drone and fielded its own version in Venezuela at $35,000 per unit — versus $1.3 million for a Tomahawk cruise missile. Understanding how things are built is not a hobby. It is a strategic capability. There is no reason it cannot exist here as public infrastructure.

I propose we establish physical teardown facilities at our ITEs and polytechnics — not as one-off workshops, but as a vertical curriculum progressing from guided disassembly to design analysis to cost modelling. We should reverse-engineer frontier hardware - EVs, batteries, solar, drones. Structured public reports should feed the industrial commons. Such facilities could also be housed at A*STAR’s centres, which already run model factories and joint industry labs. Partnered with ITEs and polytechnics, with structured access for SMEs, they would form a strong starting point.

For ordinary citizens, the National Library Board’s MakeIT at Libraries — its free maker-space programme run with IMDA, offering 3D printing, digital cutting, robotics, and coding classes in selected regional libraries — teaches the basics well. I have observed 3D printing and robotics classes there, and the Starter Sessions are well-designed. But the curriculum stops there, with no intermediate or advanced progression. If we are serious about life-long learning, MakeIT needs a vertical curriculum — Stage 2 and Stage 3 modules, project-based, progressing from operating the machines to designing with them and using them more intensely — not just a wider menu of first lessons.

Our industrial clusters are well-designed. But co-location is necessary, not sufficient. Knowledge does not diffuse simply because firms share a postcode. It diffuses through people. Tacit production knowledge — the working understanding of why a process line behaves the way it does, what failure modes to watch for, which suppliers to trust — does not transfer through papers or seminars. It transfers through people, on the floor, over months. Taiwan’s ITRI is the clearest example: between 1973 and 2008, thirteen thousand ITRI staff moved into Taiwanese industry, many of them seeding what became TSMC, UMC, and the wider Hsinchu semiconductor cluster. Engineer circulation, not patents or licensing, was the primary transmission mechanism.

To speed up the rate of knowledge diffusion, I also propose an Industrial Commons Rotation: a statutory obligation on large firms — GLCs and major anchor MNCs — to second mid-senior technical staff into SMEs within the same sectoral cluster for six- to eighteen-month rotations, with reciprocal SME-to-large-firm flows so the small firms’ problem-solving feeds back into the larger firms’ processes. We can do this as a pilot in the tacit-knowledge sectors I named earlier — semiconductor back-end, aerospace MRO, petrochemicals, marine and offshore. If we want an industrial commons, we must make the people circulate.

Teacher rotation through industry should be a statutory requirement. Polytechnic and ITE instructors currently can take industry attachments as a form of voluntary continuing professional development, but the mechanism is discretionary, not structural. I propose a pilot: twelve months in industry for every sixty months of teaching.

This is the first task: making the knowledge already inside Singapore circulate.

(Industrial commons there)

The second task is to go to where the knowledge is being created.

The Overseas Markets Immersion Programme, launched in 2024, is the right vehicle in form. In substance, it is far too small. The initial target was 250 individuals over two years; about 120 had been supported by early 2026. Salary support of up to seventy per cent for nine months, capped at $5,000 a month, with a minimum salary threshold of $4,000 a month that effectively restricts it to PMETs.

If I had to name one programme that could be reframed tomorrow at low cost and high return, it is OMIP. Three changes:

First, expand it by an order of magnitude — from a 250-person pilot to a 5,000-person annual cohort within five years.

Second, remove the minimum salary threshold for technical and operational roles in priority sectors, so ITE-trained engineers and polytechnic graduates can be deployed to cutting-edge manufacturing clusters in China, Japan, and Europe — not only PMETs to regional roles.

Third, reframe the strategic purpose. OMIP is currently designed as outbound business development support — helping Singapore companies expand into new markets. . It should also be inbound knowledge acquisition — sending Singaporeans to where manufacturing knowledge is being created, to bring it back. The architecture is broad enough to hold both, if we choose to use it that way.

Conclusion

Sir, in closing. The deepest lesson of the last decade of skills policy — here and abroad — is this: the state cannot substitute for the firm as the principal in a worker’s training. We have spent ten years and considerable public money trying. The next ten years should be spent designing the system in which sectors and firms — and the workers themselves, through their unions — are forced to care about the answer.

Sectoral co-governance with named accountability. Sectoral bargaining where tacit knowledge is deep. Engineered knowledge diffusion where it is not. And an order-of-magnitude expansion of how we send Singaporeans to where cutting-edge knowledge actually lives.

Capability is tacit knowledge. It deepens inside sectors and firms — which is why we need sectoral architecture that holds workers there long enough for it to form. And it circulates through an industrial commons — which is why we need engineers, instructors, and Singaporeans abroad moving through that commons, carrying knowledge with them. The state is an orchestrator, it builds the architecture in which both happen.

Thank you.