Speech on Motion: An Artificial Intelligence (AI) Transition with No Jobless Growth
6th May 2026
Speaker, I declare my interest as a director of a company that makes AI-enabled applications and consults on the same.
Two Moments
In the three and a half years since ChatGPT’s release, I have had two moments of awe and dread.
The first was in November 2022. GPT-3.5 could iterate on software features, generate ideas, write code. Five years ago, it was received wisdom that everyone should learn to code. Today, coding ability is cheap and abundant. Computer science graduates — even from top schools like Stanford — are finding it difficult to find jobs. GPT-2 was a toy that generated amusing limericks. Three years later, its successors made an entire profession’s scarcity disappear. We used to talk about prompt engineering in 2023 and 2024. That talk has died down too.
The second was in November 2025, when Anthropic released Claude Code — a reliable AI agent paired with a frontier model. I could leave the computer running overnight, and there would be work done at the end. It is a different experience from chatting with a chatbot. The chatbot engages you in back-and-forth — refining your ideas, indulging your whims, steel-manning your speeches. The agent, unless it needs clarification, just goes and does things. It may be off by a bit, but you give your input and it takes another 5 or 50 minutes before it comes back with the problem solved. A very smart junior colleague.
And now we have AI agents — Claude Code, Codex — tools that have made me, if I may borrow internet lingo, Claudepilled. I use Claude Code for my own work. I can give it the most wishy-washy specification, and it returns a wonderful data workflow or website layout. For someone who could never build a pretty website to save his life, it is liberating. There is a spirit of play in working with these tools that I think every Singaporean deserves to experience.
It is an exhausting world it heralds. Software engineers pulling 80-hour weeks while running multiple AI agents overnight — so that someone, human or machine, is always on the clock. Jobseekers — especially recent graduates in white-collar work — applying to hundreds of jobs without a single interview. Job portals like LinkedIn have become memory holes for resumes, where the lived experience is like shooting an application into the void.
The pace of change humbles us all. I am suspicious of any assertion that starts with “AI will never…”, because the shelf-life of those predictions tends to be measured in months.
What concerns me is not the destination, but who gets left behind on the way there, and whether we are building the institutions to ensure no one does.
I have three propositions. First, that access to premium AI — and especially to AI agents — must be universal, not gated by course enrolment or union membership.
Second, that we must treat the handful of companies building frontier AI with the same strategic seriousness we bring to bilateral relations with countries — because their decisions on pricing, access, and deployment now shape our productivity frontier as directly as any trade agreement.
Third, that we must buy time for workers by upgrading our retrenchment framework for AI-speed displacement.
I. Model Access as a Right
Sir, I believe access to premium AI — and especially AI agents — is a right, not a privilege. Intelligence, in the sense of uplift, should not stratify according to wealth. I spend a couple of hundred dollars a month on these tools because they are game-changing. But for those who cannot afford to, it bakes in inequality from the start. Does it simply disqualify them from the off?
The Government has partially adopted the 2024 suggestions of my colleague Gerald Giam to provide universal premium AI model access. The SkillsFuture premium-AI access scheme — six months of free tools for Singaporeans who enrol in selected courses - is a step in the right direction. Likewise, NTUC’s subsidies, covering 21 AI tools.
These are good starts. But they are unnecessarily gated behind course enrollment and union membership. And critically, they likely will not cover AI agents — the tier where the real productivity gap will open.
Why does this matter? AI agents are expensive to run. We may hope agent access follows the cost curve of internet bandwidth or compute — but there is no necessary reason it should. It is an empirical question.
Anthropic’s CEO said in January that 80 per cent of its revenue comes from enterprise customers, driven by API calls on a pay-per-token model. If agents remain enterprise-grade by default, then individual citizens — jobseekers, freelancers, retirees — are locked out of the tier where the real productivity gains are being made.
Three possible directions. One: negotiate sovereign access — a bulk licensing agreement with frontier AI providers for volume-discounted agent access for all citizens. Two: if agent access is employer-provisioned in the market, make it universally so — require companies above a certain size to provide agent-grade AI to all employees, the way we require CPF. Three: if frontier agents remain too costly, identify a minimal viable agentic tier and fund that universally.
Will the Government make premium AI access a universal entitlement, rather than gate it behind course completion or union membership?
II. Companies with Sovereign-Grade Consequences
Sir, I learnt recently that even AI engineers at the global top two or three frontier AI labs — are worried about falling behind because they cannot use Claude Code. And having just returned from China, I learn first-hand that one cannot use Claude Code there at all — Anthropic blocks API calls from mainland China, Hong Kong, and Macau entirely.
If even the engineers building frontier AI are desperate for access to one another’s tools — and if entire countries can be locked out — then access is not a convenience. It is a strategic capability. And the question for Singapore is whether we will secure it, or whether we will be price-takers forever.
There are perhaps three to five companies in the world whose decisions on pricing, access, and deployment will shape every economy’s AI trajectory. When Anthropic or OpenAI decides what to charge for agent-tier access, or whom to serve, that decision shapes Singapore’s productivity frontier as directly as any trade agreement.
We should therefore treat this class of companies — frontier AI firms that have crossed a threshold of systemic importance — with the same strategic seriousness we bring to bilateral relations with countries. Not because they are sovereign — they lack the durability and legitimacy of states, and remain subject to home-state law. But because their decisions carry sovereign-grade consequences for our economy, and we should engage them accordingly.
What does that mean in practice? Four things.
First, negotiate access at the sovereign level. In the possible future where frontier AI agent costs go up, not down, Singapore should seek bulk licensing agreements for agent-tier access the way we negotiate energy supply — not individual subscriptions, not course-gated subsidies. This means accepting that frontier AI access may be a permanently higher line item in national expenditure, and procuring it systematically, because the alternative — citizens priced out of the tools that define productivity — is worse.
Second, we trade based on what we have. Nvidia CEO Jensen Huang has described the AI stack as a five-layer cake: energy, chips, infrastructure, models, applications. In my view: we do not have energy at scale. We do not have frontier model capability. At the application layer, there is little moat outside of the knowledge agglomerations we can build for ourselves — we would be competing with some of the highest cost bases in the world.
But Singapore Inc builds good data centres. And we are among the world’s leaders in water reuse and integrated water management — which is a binding constraint on data centre expansion across water-scarce regions in Southeast Asia. If we position ourselves as the infrastructure partner of choice for this region, that is real leverage — something we bring to the table in exchange for access, for pricing, and for presence.
When a company like Anthropic or OpenAI approaches us, we should be their preferred regional bilateral partner in rolling out and scaling their data centre buildout regionally, as well as all the infrastructure needed to make these data centres work.
Third, attract real technological presence. We should seek frontier AI companies establishing development offices here — not predominantly sales offices, which was the experience with the FAANG companies in the 2010s. And I would prefer we be quality-conscious. Most AI companies are not frontier AI companies. We need targeted strategy and engagement with frontier AI companies specifically.
Fourth, get Singaporeans inside these labs. Once you are in the frontier AI ecosystem, it becomes much easier to circulate within that group of companies. I would welcome the Government doing some fact-finding — engaging our local and overseas Singaporeans already in these roles, understanding how they or their colleagues got hired, and disseminating that to our students and technical researchers. Right now, anecdotally, half a million to million-US-dollar salaries (excluding equity) in the US for AI researchers are fairly common, and it’s clearly in our interest to figure out how to get more Singaporeans into this tight labour market. I would really like to see the Skills Framework for frontier AI lab researcher!
III. Buying Time
Speaker, my last point is about the transition. Let me start with a person.
In Hangzhou, a quality assurance supervisor named Zhou joined a tech company in late 2022 at 25,000 yuan a month — about S$4,800 — reviewing AI model outputs for accuracy and safety. In 2025, his employer decided an AI model could do his job. They offered him a reassignment at roughly 40 per cent less pay. He refused. They terminated him. Zhou went to arbitration and won. The company sued and lost. The company appealed — and lost again, at the Hangzhou Intermediate People’s Court. The ruling was published on 28 April this year, three days before International Workers’ Day.
The court’s reasoning is worth our attention. The company argued that AI had made Zhou’s role obsolete — a “major change in objective circumstances” justifying dismissal under China’s Labor Contract Law. The court disagreed. AI adoption, it held, is a deliberate business strategy, not an unforeseeable event. A company that chooses to automate cannot unilaterally shift the full cost of that decision onto the worker. The company had not shown the contract was impossible to perform, and the reassignment at 40 per cent less pay was not a reasonable alternative. The court added that companies should prioritise retraining workers and helping them transition to higher-level roles.
The principle — that a deliberate business decision should not externalise its full cost onto the worker — deserves serious consideration in Singapore.
If a Singaporean Zhou were retrenched tomorrow under our existing framework, would he win? Our existing Tripartite Advisory on Managing Excess Manpower and Responsible Retrenchment — TAMEM — is advisory, not statutory. An employer can lawfully automate a role and terminate the worker without first attempting to redeploy or reskill them — and the public purse, through SkillsFuture and Workforce Singapore, picks up the cost of that worker’s transition. There is no AI-specific notice period. There is no statutory redeployment-first obligation. There is no individual cause of action for the worker to challenge the reason for her termination.
The data suggests we are entering the zone where this matters. MOM’s own Q4 2025 Labour Market Report records 14,490 retrenchments in 2025 — up from 12,930 the year before. PMET retrenchment incidence reached 10.1 per 1,000 resident employees — above the pre-recessionary norm of 8.0 set during 2015 to 2019. Retrenchments were concentrated in Financial Services, Information and Communications, and Professional Services — the most AI-exposed sectors. Information and Communications employment declined outright in 2025.
The Government announced the Tripartite Jobs Council on 30 April. I welcome its intent. But it creates no new powers, no new obligations on employers, and no new rights for displaced workers. How does the Government intend for this Council to work?
Too often, what workers experience is not a frank conversation about AI-driven restructuring but a Performance Improvement Plan — a process that, in many cases, is a bit of wayang designed to paper over a predetermined outcome. I foresee that such potentially misleading reasons may be given, and workers must have the power to be able to challenge this.
I propose three directions. First, a 90-day mandatory transition notice before AI-driven role elimination. Second, a redeployment-first obligation — retraining or reassignment before AI-driven termination. These provisions will slow the velocity of AI disruption. And velocity is what determines whether adjustment is possible. Third, for workers to be able to substantively challenge the reasons of their termination if they feel they are misleading, so that these AI restructuring protections will be real.
IV. Conclusion
Speaker, in closing.
Finland gave people unconditional cash as income. They were happier, less stressed — and the great majority still walked into the employment office and asked for work. The American pollster David Shor polled Americans this year: three to one, across every political persuasion, they chose job creation over direct transfers.
People, when offered the choice between a universal basic income and employment, invariably choose employment. Not because they are irrational. Because a job is where you are needed — and being needed is not something a universal basic income can replace.
So no jobless growth — yes. But more than that: no growth where the gains are captured disproportionately by capital and the burden of adjustment falls on labour. Universal access, so intelligence is not rationed by wealth. Strategic engagement, so we are not price-takers in our own future. And a retrenchment framework where the company that decides to automate bears the cost of that decision before the worker does.
I do not think the awe and dread goes away. But in a country that builds for its workers, there is hope for a brighter future.
Thank you.