From 50 People to 5,700,000 — One AI-native Architecture, Two Scales

This piece puts two kinds of experiments side by side: “companies built AI-native from day one” and “Singapore as an AI-native nation”. The first is the 50–500-person sample described in YC’s 2026 batch “How To Build A Company With AI From The Ground Up”; the second is the 5.7-million-person sample described in Budget 2026. Same architecture, very different levers.

I. One Architecture, Two Scales

In 2026, two kinds of AI-native experiments are running at the same time.

One is the 50–500-person company — Anthropic, Cursor, Lovable — putting AI at the core of the workflow from day one. YC partner Diana puts the methodology most precisely:

“It should not be a tool your company just uses. It should be the operating system your company runs on.”

The other is a 5.7-million-person city-state. Budget 2026 puts AI at the strategic core of the entire budget; PM Wong personally chairs the National AI Council; the Committee of Supply across 11 ministries is built around AI — the only sovereign state to date to have explicitly named “AI-native nation” as a strategic goal.

Put the two side by side, and an overlooked fact appears: AI-native is not a matter of scale — it is an architecture.

Three data signals — why Budget 2026 is real

• In Lawrence Wong’s Budget 2026 closing speech, AI is, for the first time, elevated to the strategic spine of the entire Budget
• RIE2030 funding S$37B (announced 2025-12, in effect 2026-04)
• EDB has landed >S$30B in foreign data centre investment (Microsoft S$5.5B, AWS S$12B, Google US$9B)
• A coordinated AI debate across 11 ministries’ COS (MDDI / MTI / MOH×3 / MOE / MOM / MOT / MND / MSE×3 / MSF×2 / MCCY)

Operational definition

What does it mean to call an organisation “AI-native”? Three tests:

1. AI sits on the critical decision path: core judgements default to AI first, with humans approving exceptions
2. Workflows assume AI by default: processes, data, and artifacts are designed assuming AI is the primary consumer
3. Incentives align with AI augmentation: promotion, hiring, and training do not reward “human heroes who route around AI”

In a 50-person company these three can be redesigned within a few months; on a 5.7-million-person nation, doing them through the civil service, the private sector, and universal education — 5–10 years is an optimistic estimate.

A company satisfies these three in one way; a country must satisfy the same three in a completely different way. Singapore’s bet is not on compute / data / models — it has no edge on those three factors of production. It is betting on execution architecture: using national-level levers to amplify the speed of the AI-native transformation of its homegrown enterprises. A country cannot be “AI-native” by itself — it has to penetrate into enterprises. That is the core argument of this piece.

II. What an AI-native Company Looks Like

YC’s 2026 batch “How To Build A Company With AI From The Ground Up” lays out the methodology cleanly. This section borrows Diana’s five pillars directly, then adds Jack Dorsey’s “kill the middle layer” organisational design at Block — together a complete, operational checklist for “how a 50-person company becomes AI-native”.

The next section then maps the same structure onto the national scale.

1. Closed Loop

This is the load-bearing one. Diana uses the cybernetics concept:

• Open loop: decide → execute → don’t measure systematically → don’t adjust. Inherently “lossy”.
• Closed loop: a self-regulating system that continuously monitors output, adjusts the process, and runs more accurately over time.

Old-world companies are basically open loop: decisions get made, work gets done, results may or may not be measured systematically, and the process never self-adjusts. The first principle of an AI-native company is to turn every important workflow into a closed loop — artifacts go to the AI, the AI sees the full context, and it adjusts the next step automatically.

2. Queryable Organization

For the closed loop to run, the organisation has to be queryable by AI. In practice:

• All meetings recorded by an AI notetaker
• Reduce DMs and email; embed agents in every communication channel
• Build dashboards that pull in all data — revenue, sales, engineering, hiring, ops — everything
• Give agents access to Linear, Slack, Pylon, GitHub, Notion, sales call recordings, daily standup

The core principle: to get the model’s full capability, you must give it the same order of magnitude of context as your employees have.

3. AI Software Factories

Software development gets rewritten:

• Humans: write the spec, write the test suite that defines “success”
• Agent: generates implementation code, iterates until tests pass
• Humans: judge whether the output is acceptable

The extreme form: the repository contains no hand-written code, only specs and tests. Strong Compute’s AI team works this way — they let agents iterate until they hit a probabilistic satisfaction threshold, with the goal of eliminating “humans writing code or reviewing code”.

4. Kill middle management / three employee archetypes

The old world needed middle managers to route information up and down the org. Inside an AI-native company, the intelligence layer replaces that function — “there should be almost no human middleware”.

Jack Dorsey at Block:

“If you keep the same org chart and management structure, you’ve missed the shift entirely. The company itself has to be rebuilt as an intelligence layer with humans at the edge guiding it rather than routing information through it.”

The future company has only three roles:

5. Token-maxing replaces headcount

Resourcing gets rewritten:

• One person using AI tools = an entire engineering team of the old era
• Engineering, design, HR, admin should all shrink dramatically
• You should be willing to tolerate “uncomfortably high” API bills — they replace far more expensive, far more bloated headcount costs
• The best companies will be token-maxing companies

Bridge

These five rules can each be redesigned in a few months inside a 50-person company.

What about a 5.7-million-person nation?

III. What an AI-native Nation Looks Like

Budget 2026 is currently the only complete answer. This section starts with the top-level narrative, then expands all the ministry-level deliverables under six “levers”, and closes with three layers of risk management and a transparency paradox.

3.1 Top-level narrative: from a departmental issue to a strategic spine

Signal 1 — Political narrative is upgraded

In his Budget closing speech on 6 March 2026, Lawrence Wong for the first time elevated AI to the strategic spine of the entire Budget, and positioned the entire Budget as Singapore’s national action plan for a world that is “more contested, more fragmented and ultimately, more dangerous” — with AI as a key strategic chip.

Signal 2 — Organisational level upgraded

PM Wong personally chairs the National AI Council — not delegated to MDDI alone. This is Singapore’s signature move when an issue is identified as “national-level”: a critical issue is not handed to a single ministry, but parked in the Prime Minister’s Office.

A coordinated AI-themed Committee of Supply debate across 11 ministries is the most concentrated AI moment in the history of Singapore’s COS.

Signal 3 — Fiscal level upgraded

• RIE2030 funding S$37B (announced 2025-12, in effect 2026-04)
• Public AI research investment S$1B+ (2026–2030)
• EIS 400% tax deduction extended to AI (YA 2027–28, S$50K/firm/year cap)

These three signals together = the national-level AI Founder model is in place. Josephine Teo single-handedly coordinates MDDI, IMDA, and the international AI governance track — across the 57 official ministerial speeches we could find, she gave 23.

3.2 Six levers — the full map of the national AI-native effort

We sort all AI-related Budget 2026 + ministry + statutory board policies and deliverables along the “AI introduction pathway” into six levers: Infrastructure, Governance, Talent, Applications, Government adoption, Diplomacy.

This is more revealing than sorting by department (MDDI / IMDA / MAS / MOH / …) — readers can see the overall shape at a glance. Each lever cuts across multiple ministries; only when stitched together does the full execution pipeline appear.

Lever 1 — Infrastructure (data + compute + physical)

What the state does directly: build the compute and data substrate that enterprises cannot afford to build themselves.

Foreign-invested compute (EDB):

• Microsoft data centre S$5.5B
• AWS S$12B
• Google US$9B / S$11.6B + DeepMind Lab
• NVIDIA × SIT Centre for AI, × Singtel, × AI Accelerator
• OpenAI Singapore APAC regional HQ
• Anthropic recruiting Singapore Country Lead (2026)

Domestically subsidised compute:

• Enterprise Compute Initiative (ECI) S$150M — direct compute subsidies for enterprises
• one-north AI Park / Kampong AI (MOF)

Capital platforms:

• Anchor Fund @ 65 second tranche S$1.5B
• Future Sectors Development Fund S$1.5B
• EQDP expanded to S$6.5B

National data substrates:

• MOH/Synapxe HEALIX = national healthcare data + AI infrastructure
• URA Virtual Singapore = national-scale digital twin
• BCA BETC Grant S$100M = digital infrastructure for the construction industry
• JTC Punggol Digital District + Open Digital Platform (ODP) = a fully smart district

Homeland-security-side compute:

• HTX NGINE — NVIDIA B200 DGX SuperPOD (homeland security’s own compute)

Lever 2 — Governance (rules + sandboxes + law)

What the state does directly: make enterprises dare to deploy. The biggest reason firms don’t deploy AI isn’t technology — it’s compliance risk.

General governance frameworks (IMDA):

• Model AI Governance Framework (2019)
• AI Verify (2022) + AI Verify Sandbox (10+ MNCs participating)
• GenAI Eval Sandbox + GenAI Sandbox 2.0
• Generative AI governance framework (2024)
• Agentic AI Governance Framework (2026-01-22 at Davos, world-first)
• Trusted Data Sharing Framework + DPTM upgrade SS 714:2025

Five-layer governance stack for finance (MAS):

• FEAT Principles (Fairness / Ethics / Accountability / Transparency)
• Veritas Initiative v1 / v2 / v3
• Project MindForge (24 institutions + Microsoft / AWS / Google / Nvidia)
• AI Risk Management Guidelines
• BuildFin.ai

Cybersecurity governance (CSA):

• Securing AI Systems Guidelines + Companion Guide
• Securing Agentic AI supplement
• Frontier AI Risk Advisory
• Cyber Trust Mark — AI security dimension

Legal governance (MINLAW + IPOS):

• Copyright Act §244 = AI training exemption (tied with Japan as the most permissive in the world)
• IPOS “When Code Creates” report — AI authorship position
• Strict on outputs: OCHA + Elections Bill 2024 (deepfake ban) + Criminal Law Bill 2025 (AI-generated intimate imagery criminalised) + Online Safety (Relief and Accountability) Bill 2025

Governance philosophy: permissive on training, strict on outputs. Japan and Singapore are currently the only two countries in the world that have done this — it gives enterprises a clear, predictable boundary.

Lever 3 — Talent (education + training + transition)

What the state does directly: give enterprises access to people who can actually use AI.

Mass-public layer (MDDI):

• AI Bilingual 100K Workers programme (first cohort accountancy + legal, going live 1H 2026, in partnership with ISCA / SAL / SCCA)
• National AI Literacy Programme

Professional layer (IMDA + AISG):

• TechSkills Accelerator (TeSA) AI extension
• AISG AI Apprenticeship Programme (AIAP): 16 cohorts, 410+ apprentices, 900+ applicants, new cohort 800 places
• AISG 100E Programme (S$150K joint investment per project)

Education-system layer (MOE + NIE):

• SLS (Student Learning Space) AI tool stack with 8 categories
• GenAI Use Guidelines + AI Ethics Framework
• EdTech Masterplan 2030
• NIE AI@NIE + Certificate in AI for Education
• Microsoft Elevate × Singapore (AI in higher education)
• NUS / NTU / SMU / SUTD comprehensive AI curriculum reform

Fiscal subsidy layer (SSG + WSG):

• SkillsFuture AI courses 50% / 70% tiered subsidy
• Mid-Career S$4,000 Credit
• SkillsFuture Level-Up Programme
• WSG × SSG merger = one-stop skills-and-employment platform

Mid-career retraining layer (MOM):

• Job Redesign+
• Career Conversion Programme (CCP)
• Enterprise Workforce Transformation Package (EWTP)
• NTUC × AI worker protections

“Not all of us can be AI engineers. But we can be ‘bilingual’ in AI in our own areas of expertise.”

— Josephine Teo, MDDI Committee of Supply, 2026-03-02

Lever 4 — Applications (industry + public service deployment)

What the state does directly: roll out flagship applications simultaneously across 11 ministries.

Industry flagships (MTI):

• National AI Missions (4 priority sectors)
• AI Centres of Excellence
• Embodied AI R&D
• AI upgrades inside the Industry Transformation Maps (ITM)

Research flagships (A*STAR):

• A*STAR CFAR five research pillars
• AI Manufacturing 2030 (Mencast propellers)
• AI materials screening 50–100x acceleration
• GIS + SingHealth health-AI partnership
• National Multimodal LLM Programme S$70M

Regional LLM flagship (AISG):

• SEA-LION v3 / v4 / Guard
• SEALD (datasets)

Enterprise diffusion (IMDA + ESG):

• NAIIP — National AI Impact Programme: 10K firms + 100K workers / 2026–2029
• Champions of AI (flagship enterprise programme)
• ESG PSG AI subsidy ratio 30% → 50%
• ESG SMEs Go Digital AI module

Healthcare (MOH + Synapxe):

• Note Buddy — GenAI clinical-note assistant (5,000+ clinicians, 67K records as of 2025-12)
• HealthHub AI (citizen-facing, 4.5/5 rating)
• AimSG (national medical-imaging AI)
• SELENA+ (diabetic retinopathy screening)
• ACE-AI (chronic-disease risk prediction, expanding to all ~1,100 Healthier SG clinics in early 2027)
• APOLLO (national CT coronary AI)
• Healthier SG × digital twin (chronic kidney disease management)

“AI-enhanced, not AI-decided — clinicians remain in the loop.”

— Ong Ye Kung, MOH Committee of Supply, 2026-03-05

Transport (MOT + LTA + PSA + CAG):

• Punggol AV public shuttle (first commercial AVs, 3 routes live 2025-12)
• CETRAN AV national test centre
• PSA Tuas Mega Port = world’s largest fully automated port by the 2040s
• Changi the world’s first ISO/IEC 42001 AI governance certification

Construction and urban (MND + HDB + BCA + URA + JTC):

• Built Environment AI Centre of Excellence (BE AI CoE S$30M)
• BCA Integrated Digital Delivery (IDD)
• SPRINT programme — green-channel public procurement of construction AI
• HDB Tengah = first smart-energy town with 42,000 units
• HDB AskJudy + MSO OneService

Environment and water (MSE + NEA + PUB):

• NEA Weather Science Research Programme S$25M
• Dengue AI prediction + mosquito-vector control
• PUB Smart Water Meter Programme + Joint Operations Centre + Bentley leak detection

Lever 5 — Government adoption (procurement / lead by example)

What the state does directly: get civil servants using AI first — to set a precedent for enterprises.

Civil government (GovTech):

• Pair (civil service AI assistant, 150K civil servants target)
• Pair Search (Hansard + courts + legislation searchable)
• LaunchPad (3K MAU / 400 ideas)
• AI Trailblazers 1.0 + 2.0
• Litmus + Sentinel (AI safety toolkit)
• Agentspace = Asia’s first air-gapped agentic AI

Defence (MINDEF + DSTA + DSO + DIS):

• DIS — SAF Digital and Intelligence Service (fourth service established 2022, restructured into DCCOM + SAFC4DC in 2025) = AI written into the structure of the service itself
• DIS × AI Singapore MoU + DIS Sentinel Programme
• DSTA × Shield AI (autonomous drones) + Thales AI Co-Lab + Anduril Lattice
• DSTA × RSN computer-vision ship classification
• DSTA in-house GenAI tools + DSTA × MIT CSAIL
• DSO × Mistral AI defence GenAI
• DSO × Alan Turing Institute MoU

Homeland security (HTX + SPF + ICA):

• HTX Phoenix LLM (in-house trained)
• HTX H2RC humanoid robotics centre S$100M (Q2 2026 launch)
• HTX × Google Cloud / Microsoft / Mistral AI / Firmus / Singtel / ST Engineering
• SPF Anti-Scam Centre / Anti-Scam Command — RPA + AI
• SPF PolCam + GIBSON airport robot + Smart Glasses real-time video analytics
• ICA Multi-Modal Biometrics System (MMBS) — iris + face

Lever 6 — Diplomacy (international governance + FDI + standards)

What the state does directly: get foreign firms to put their AI governance HQs in Singapore.

This is the only way 5.7 million people can lever G7-level voice.

Singapore-led global frameworks:

• Singapore AI Safety Institute (AISI) — S$10M/yr
• Singapore Conference on AI / International Scientific Exchange on AI Safety I + II
• Singapore Consensus on Global AI Safety Research Priorities (signed by 11 countries, including the US and China)
• IMDA × Humane Intelligence multilingual red-team challenge

ASEAN regional:

• ASEAN Working Group on AI Governance (WG-AI)
• ASEAN Guide on AI Governance and Ethics (adopted by 10 countries)
• ASEAN Hanoi Declaration 2026 (digital ministers’ meeting)

Bilateral cooperation:

• US-Singapore Smart Cities Programme + Digital Economic Cooperation Roadmap
• ROK bilateral AI cooperation
• EU-ASEAN AI governance dialogue

Military / security:

• REAIM Asia Regional Consultations (Singapore as co-chair)
• REAIM Seoul Summit 2024 (Singapore as co-host)
• Bletchley Park / Seoul / Paris AI Safety Summits — full participation

UN + global:

• UN Global Dialogue on AI Governance + Independent International Scientific Panel
• AI Singapore × UNDP global AI literacy

Using 0.07% of the world’s population to claim G7-level voice in AI governance — this is the most non-replicable part of Singapore’s strategy.

3.3 Three layers of risk management

National-level AI-native must run risk management at the same time — something that’s far simpler at company scale, and gets very complex at country scale.

Economic risk layer — political pressure from the PMET middle class (the biggest variable)

“AI is a gamechanger. It can augment workers or displace them, depending on how work and jobs are redesigned.”

— Tan See Leng, MOM Committee of Supply, 2026-03-03

Tan See Leng’s line is not generic; it is direct reassurance to a core electorate. MOM repeatedly emphasises “mid-career PMEs face highest risk” + “job redesign for human-with-AI”.

This is a political variable that did not exist in the Smart Nation era. In the Smart Nation era, unemployment risk fell on blue-collar workers and entry-level clerks. In the AI era, the first cut goes at the PMET middle class — junior lawyers, junior accountants, junior analysts, junior engineers. Singapore’s political stability rests, to a significant degree, on the sense of security of the PMET middle class.

Risk: this could give rise to laws restricting AI’s substitution for human labour — which would unwind the entire strategy.

Social risk layer — protecting vulnerable groups

Concerns surfaced in MSF and MCCY COS debates:

• AI deepfake sexual exploitation threatens children and vulnerable groups (Rachel Ong, MSF COS 2026-03-05)
• AI automation displacing traditional roles for people with disabilities — packing, sorting, basic admin, coding (Neo Kok Beng)
• Online Safety Commission Phase 1 covering child-image abuse
• ECDA Inclusive Support Programme (InSP)
• AI economic readiness in the Malay/Muslim community (Saktiandi Supaat, MCCY COS)

Security risk layer — critical infrastructure + national security

• CSA Securing AI Systems Guidelines + Frontier AI Risk Advisory
• DIS / DSO / SPF / HTX internal AI deployment is not publicly disclosed
• The non-regulatory stance on AI chatbots used in adolescent mental-health counselling (Koh Poh Koon in oral-answer-4051): the government considers tracking infeasible and is instead promoting legitimate alternatives (mindline 1771, mindline.sg, CHAT) + app-store age verification

3.4 The transparency paradox

A pattern that surfaced during the research, worth flagging on its own because it is a potential fault line in Singapore’s strategy:

Singapore’s strategy stands on “execution advantage” — but the execution detail is not auditable.

In researching this piece, we found that many Budget 2026 AI projects have no single official aggregate number and have to be assembled from multiple press releases:

• Champions of AI: specific dollar figure not disclosed
• Anthropic × EDB MOU: details not disclosed
• Kampong AI: investment amount not separately listed
• NAIIP per-firm grant amount (officially “to be announced 1H 2026”)
• RIE2030 internal AI sub-allocations not separately listed

On the defence and security side it goes further: actual DIS headcount, DSO AI weaponisation research, the SPF PolCam video analytics stack — not publicly disclosed at all. AISI funding sources, whether HTX’s H2RC S$100M includes operating costs — all blurry.

When a country names “becoming a global AI governance hub” as its strategy, it must hold itself to a higher auditability standard than a company. Singapore is currently moving the opposite way — this is the strategy’s greatest long-term reputational risk.

IV. The Nation as the “Wrapper Layer” for Enterprises — Seven Transmission Levers

Re-arrange the six levers from Section III — sort them instead by “which enterprise bottleneck does it solve?” — and the overlooked fact appears:

The essence of Singapore’s strategy is to use the entire country as the “wrapper layer” of its enterprises’ AI-native transformation.

A country cannot be “AI-native” by itself — government departments are only a small share of GDP. For a country to be called AI-native, its enterprise base has to be AI-native. The real bet of Budget 2026 is to use national-level levers to amplify the speed of enterprises’ AI-native transformation.

Seven transmission levers — re-slicing the six

Key observation

Of these seven, only #6 (government adoption) and #7 (international diplomacy) are things the state does directly. The other five are the state penetrating into enterprises.

Why this argument is distinctive:

• Most articles on national AI strategy analyse “the state” and “enterprises” in parallel — and miss the nested relationship
• The elegance of Singapore’s strategy: the entire nation is a wrapper around enterprises. The country itself doesn’t need to become AI-native — it just needs to amplify enterprise transformation speed
• This explains why Singapore can be a contender despite having no compute / no data / no models — it’s betting on execution architecture, not factors of production

The two voices, isomorphic

Place YC’s Diana and Mariam Jaafar side by side, and an unexpected isomorphism appears:

“It should not be a tool your company just uses. It should be the operating system your company runs on.”

— YC Diana, 2026

“If healthcare is truly a national AI mission, the goal cannot be incremental adoption.”

— Mariam Jaafar, MOH COS 2026-03-04

The company-version line and the country-version line are saying the same thing. That is this piece’s strongest “two-pole isomorphism” evidence.

V. Closing: Three Open Questions + Three KPIs

Three open questions

1. When tools like Cursor make one engineer worth a thousand, is the national “penetrate-into-enterprise” lever still in time? Or do enterprises outrun the state?
2. Does Singapore’s “wrapper layer” bet hold — is a positioning of “can’t be AI-native by itself, but is the world’s best incubator for AI-native enterprises” enough?
3. How will the asymmetry of speed converge? AI slows, the state accelerates, or society stays out of step — which one is Singapore betting on?

Three KPIs worth tracking

• NAIIP delivery rate: by 2029, completion against the 10K firms + 100K workers target
• Diffusion speed of Note Buddy-class projects: how long to scale from 5,000 clinicians to every doctor in the country — the best measurement point for “AI-native penetrating to the front line” at the national scale
• Influence expansion of the Singapore Consensus: can it grow from 11 countries to G20? Can AISI become AI’s IAEA?

One-line ending

AI-native is not a matter of scale; it is an architecture. Same architecture, two scales, experimented with simultaneously.

Singapore is not betting on compute, not betting on models — it’s betting on the organisational-architecture innovation of the nation as the wrapper layer for enterprises. If the bet pays, what it gets is the most scarce asset of the next 20 years: the actual coordinates of global AI governance and incubation.

Further Reading

• From Digital to AI: Singapore’s Second National Transformation — Budget 2026 and Smart Nation 2014 placed side by side
• Where Is Singapore’s AI Strait of Malacca? — The rise and erosion of the AI refining-hub strategy
• Singapore AI Policy Evolution Map — A five-stage timeline from Smart Nation 2014 to the all-in AI strategy of 2026
• Parliamentary AI Focus — Full Committee of Supply debate transcripts across ministries during Budget 2026
• Policy Documents — Official Budget 2026 documents and Chinese translations