Uneven Horizons-Global AI Adoption Index: Implications for Growth, Work, and Human Experience

By Dr. Luke Soon

Introduction: AI’s Unprecedented Diffusion

As of November 23, 2025, the landscape of AI adoption continues to evolve rapidly, with new reports providingfresh insights into geographic, sectoral, and economic patterns. Building on the Anthropic Economic IndexReport (September 2025), which draws from over 1 million anonymized Claude.ai conversations and API trafficfrom August 4–11, 2025, this updated analysis integrates additional key reports released in late 2025. Theseinclude the McKinsey Global Survey on the State of AI (November 2025), PwC’s AI Predictions 2026 (October2025), the State of AI Report 2025 by Nathan Benaich and Ian Hogarth (October 2025), and the OECD AI PolicyObservatory Update (September 2025). These sources enrich our understanding by adding enterprise-levelsurveys, economic forecasts, research benchmarks, and policy perspectives—complementing Anthropic’s usagedata with broader surveys and projections.The McKinsey survey, based on responses from 1,500+ executives across industries, highlights acceleratingenterprise adoption and ROI. PwC’s predictions focus on job impacts and economic value, drawing from globalCEO surveys. The State of AI Report synthesizes research trends and industry shifts, while OECD emphasizespolicy and inequality. Together with Stanford’s AI Index 2025 and WEF’s Future of Jobs Report 2025 (alreadyreferenced), they reveal a consensus: AI drives productivity but risks deepening divides, with adoption surging20–40% year-over-year.

Technological history is punctuated by waves of adoption that reshape the economy and society. Electricity, the internal combustion engine, and the personal computer each defined eras of transformation. Yet none spread with the velocity of artificial intelligence.

According to Gallup (2025), 40% of employees in the United States now report using AI at work—a doubling in just two years from 20% in 2023 . This trajectory dwarfs the adoption of the internet, which took five years to achieve similar penetration, and personal computers, which needed two decades to become household staples.

The reason lies in AI’s accessibility: it requires no specialised training, no new hardware infrastructure beyond cloud access, and can be activated by natural language. AI’s diffusion curve has already exceeded benchmarks of prior general-purpose technologies (Stanford HAI, 2025).

Yet the Anthropic Economic Index 2025 warns: adoption is not evenly distributed. Early use is concentrated in affluent economies, tech-centric sectors, and a narrow set of tasks—particularly coding, knowledge work, and administrative automation . This concentration risks widening global and enterprise inequalities, echoing historical patterns of divergence in the 20th century when electrification and industrialisation transformed some economies while leaving others behind.

This essay explores these patterns, weaving together the Anthropic Index with insights from Stanford HAI, WEF, PwC, McKinsey, and global case studies. Through a Human Experience (HX = CX + EX) lens, I argue that AI’s diffusion presents both a profound opportunity and a systemic risk. The outcome depends on whether adoption amplifies existing divides or fosters convergence through equitable deployment.

Claude.ai’s global footprint is dominated by a handful of countries, reflecting broader AI adoption trends. The United States leads with a 21.6% share of total usage, followed by India (7.2%), Brazil (3.7%), Japan (3.7%), and South Korea (3.7%). Smaller shares go to the United Kingdom (3.2%), Germany (2.6%), France (2.2%), Canada (2.1%), and Australia (1.9%). This raw distribution skews toward populous nations, but adjusting for working-age population (15–64 years) via the Anthropic AI Usage Index (AUI) uncovers deeper insights.The AUI measures over- or under-representation relative to population: values >1 indicate higher-than-expected usage. Top performers include Israel (7.00), Singapore (4.57), Australia (4.10), New Zealand (4.05), South Korea (3.73), the United States (3.62), Canada (2.91), the United Kingdom (2.67), Luxembourg (2.74), and Switzerland (2.81). Emerging economies lag: Indonesia (0.36), India (0.27), Nigeria (0.20), and Bolivia (0.48). Globally, AUI tiers show Leading (top 25%: 1.84–7.00, e.g., Israel, Singapore), Upper Middle (0.89–1.71, e.g., Czechia, Austria), Lower Middle (0.37–0.85, e.g., Peru, Colombia), Emerging (0.01–0.36, e.g., Indonesia, Ghana), and Minimal (0.00, e.g., Aruba, Tonga).A strong power-law relationship ties AUI to income: AUI ~ GDP per working-age capita^0.69 (R²=0.709, p<0.001). A 1% GDP per capita rise correlates with a 0.7% AUI increase, suggesting wealthier nations leverage infrastructure and knowledge work for faster adoption.Stanford’s 2025 AI Index echoes this unevenness: organisational AI use reached 78% globally (up 23 percentage points from 2023), with generative AI at 71% (up 38pp). Regional growth varies: Greater China up 27pp, Europe up 23pp. Public optimism stands at 55% (up from 52% in 2022), with 66% expecting life changes in 3–5 years (up 6%). However, trust in data privacy dipped to 47% (from 50%). Investments surged to $252.3B (up 44.5%), but skewed toward high-income regions.WEF’s Future of Jobs Report complements this, projecting AI as transformative for 86% of organizations, with 93% planning AI strategies. Adoption barriers include skills gaps (63%), yet 83–91% plan reskilling for AI. Globally, 79–77% hire for AI design skills, 75–69% for AI work-alongside skills.

Metric	Anthropic (Claude.ai)	Stanford AI Index	WEF Future of Jobs
Global Adoption	Concentrated (US 21.6%)	78% organizations (up 23pp)	86% transformative
AUI/Exposure Correlation	~GDP^0.69 (R²=0.709)	Investment $252.3B (up 44.5%)	Skills disruption 75–95%
Optimism/Strategies	N/A	55% beneficial	93% AI strategies

Part I. The Speed and Nature of AI Adoption

The doubling effect

The doubling of workplace AI use in two years signals not just hype but real integration into workflows. Gallup’s data mirrors PwC’s AI Jobs Barometer 2025, which finds 3 in 4 jobs already show measurable AI exposure. Exposure here does not mean displacement; rather, it reflects augmentation, automation, or redefinition of tasks.

Stanford HAI’s AI Index 2025 contextualises this: AI adoption is faster than any general-purpose technology in modern history, driven by three accelerants:

Digital infrastructure readiness – AI rides on the backbone of cloud and mobile.
Ease of interface – natural language replaces coding as the medium.
Frontier model improvements – each release expands viable tasks.

Adoption vs. capability

The Anthropic Index data show an intriguing shift. Educational and scientific tasks rose in share from 9% to 12% and 6% to 7% respectively in 2025 . Meanwhile, business operations and management tasks declined. This suggests AI is diffusing fastest into domains of knowledge synthesis and explanation rather than traditional managerial work.

Case Study – Higher Education:

The University of Melbourne deployed AI tutors to augment staff in its School of Engineering. Student feedback indicates a 15% increase in learning satisfaction and better accessibility for international students. Faculty report reduced cognitive load in grading, freeing more time for mentorship.

Case Study – Financial Services:

A North American bank adopted AI agents for compliance monitoring. By automating suspicious activity report (SAR) drafting, it reduced cycle time by 60% and reallocated analysts to judgement-intensive investigations.

Part II. Geography of Adoption: Winners and Laggards

The per capita story

The Anthropic Economic Index shows that small, innovation-led economies dominate AI usage per capita: Israel (AUI 7.0), Singapore (4.57), Australia (4.10), New Zealand (4.05), South Korea (3.73). Large economies like the US (3.62) and UK (2.67) follow, while France, Japan, and Germany lag near ~1.9. Emerging economies such as India (0.27) and Nigeria (0.20) show low per-capita adoption but large absolute volume .

Key insight: High-AUI countries tend toward augmentation (human–AI collaboration), while low-AUI economies lean on automation, even when task mix is controlled .

The Anthropic AI Usage Index (AUI) reveals striking disparities :

Leaders: Israel (AUI 7.0), Singapore (4.6), Australia (4.1), South Korea (3.7).
Moderates: US (3.6), UK (2.7), Canada (2.9).
Laggards: India (0.27), Indonesia (0.36), Nigeria (0.2).

In absolute volume, the US dominates (21.6% of global use), followed by India (7.2%) and Brazil (3.7%). But when adjusted for working-age population, small innovation-driven economies leapfrog.

Small, innovation-led economies dominate per working-age capita adoption: Israel (AUI 7.00), Singapore (4.57), Australia (4.10), New Zealand (4.05), South Korea (3.73). The US sits at 3.62; the UK 2.67; Canada 2.91. Major economies like France (1.94), Japan (1.86), and Germany (1.84) trail their innovation peers. At the other end: Bolivia (0.48), Indonesia (0.36), India (0.27), Nigeria (0.20).

AUI scales with prosperity: a power-law fit shows AUI ~ GDP^0.69; a 1% rise in GDP per working-age capita associates with a 0.7% rise in AUI (p<0.001). This reflects digital infrastructure, knowledge-worker intensity, and supportive regulation.

Interpretation for HX: higher AUI countries aren’t just “using more AI”; they are using it differently—with a tilt toward augmentation, not automation, even after controlling for task mix.

Research, Talent

From the HAI AI Index 2025, we see global divergence:

China leads in AI publications and public investment ($47.5B semiconductor fund), building on “Big Fund” support . US dominates in private AI investment and foundation models, but state-level legislation is now outpacing federal action (131 AI-related laws passed in 2024, up from 49 in 2023) . EU formalised the AI Act and launched its AI Office, focusing on risk-tiered governance and cross-border enforcement . Singapore committed $1B over five years to talent, compute, and industry adoption . Canada ($2.4B), France (€109B), India ($1.25B), and Saudi Arabia ($100B via Project Transcendence) all announced landmark AI packages .

HX lens: Countries that blend policy guardrails (EU, Singapore) with investment scale (China, US, Saudi) will shape not only economic competitiveness but also the lived experience of AI in work and society.

Governance and Safety Readiness

The FLI AI Safety Index (Summer 2025) grades seven frontier firms. It warns of a widening gap: capabilities accelerate faster than risk management.

Only 3 of 7 firms conduct substantive dangerous capability tests. Chinese firms (Zhipu.AI, DeepSeek) scored failing grades, reflecting weaker norms of self-governance despite China’s state regulatory scaffolding . US/UK firms (Anthropic, OpenAI, DeepMind) show stronger disclosure but patchy whistleblowing and third-party evals .

Country-level implication: Governance gaps are becoming a competitive differentiator. Singapore’s AI Verify, the EU’s AI Office, and US state laws position these jurisdictions as safety innovators, not just adopters.

Divergence risk

This concentration risks replicating what economist Lant Pritchett (1997) termed “divergence, big time.” Advanced economies with strong digital infrastructure and high-skill workforces can capitalise quickly, while emerging economies risk marginalisation.

The WEF’s Future of Jobs Report 2025 echoes this: without targeted investment in infrastructure and reskilling, AI could entrench inequalities.

Case Study – Singapore:

Singapore’s AI Verify, the world’s first AI governance testing framework, enables firms to pilot responsibly while gaining regulatory clarity. This governance-by-design approach helps explain Singapore’s per capita leadership in adoption.

Case Study – Brazilian Judiciary:

Brazil pioneered AI in judicial translation and case triage. The Supreme Federal Court processes thousands of documents daily using AI, reducing backlog by 25%. Adoption reflects both linguistic diversity and systemic need.

Case Study – India’s IT sector:

Despite low AUI, India dominates in absolute developer usage. Over 50% of Indian usage is coding tasks . This narrow concentration highlights how AI diffusion is filtered by local industry structure.

Israel — AUI 7.00 (leader)

A compact, innovation-dense economy topping global per-capita Claude use. High AUI aligns with strong GII standings and a workforce skewed to knowledge tasks—conditions that favour collaborative (augmentative) use rather than blunt automation. HX implication: invest in human-in-the-loop design to amplify expert judgement.

Singapore — AUI 4.57 (leader)

Per-capita usage rivals frontier hubs. The Index explicitly links high adoption to economies with robust digital infrastructure and innovation-friendly policy—exactly Singapore’s stance; it also ranks alongside Israel on innovation indices. HX implication: governance-by-design (e.g., AI assurance) sustains adoption while protecting experience equity.

Australia / New Zealand — AUI 4.10 / 4.05 (leaders)

Oceania punches above its weight. High AUI reflects cloud accessibility and knowledge-work share. HX implication: codify augmentation norms (pair-programming with models, decision trails) across professional services and public sector.

South Korea — AUI 3.73 (leader)

Advanced infrastructure and tech maturity drive usage. HX implication: combine model-centric software engineering with worker retraining in manufacturing-adjacent roles to shift from rote automation to creative augmentation.

United States — AUI 3.62 (leader)

Top-tier per-capita adoption; at the sub-national level, DC (3.82) and Utah (3.78) lead, followed by California (2.13), New York (1.58), and Virginia (1.57). This gradient suggests policy proximity, tech clusters, and industry mix matter beyond income. HX implication: design differentiated augmentation strategies by state/sector.

United Kingdom — AUI 2.67 (upper tier)

Strong but behind smaller leaders. HX implication: focus on domain-specific copilots in legal, media, and health, backed by safety assurance to close the augmentation gap.

Canada — AUI 2.91 (upper tier)

Elevated adoption with room to climb. HX implication: scale augmentation patterns from AI research hubs into resource and public services sectors.

France / Japan / Germany — AUI ≈ 1.9 / 1.86 / 1.84 (mid-tier)

Large, advanced economies lag per capita. HX implication: unlock usage by lowering organisational friction (procurement, data access), and target augmentation first in regulated professions.

Brazil — pattern outlier

Over-indexed on translation (6.4× global average) and legal drafting (5.0×)—consistent with early judicial AI adoption. HX implication: centre HX on clarity, recourse, and explainability in civic/legal interfaces.

Vietnam — pattern outlier

Over-representation in software and education tasks (e.g., cross-platform debugging 1.7–1.9×). HX implication: invest in teacher-and-coder augmentation kits to compound skill formation.

India — pattern outlier

Usage concentrates almost exclusively on software development (multiple clusters ~2.1–2.4×). Despite low per-capita AUI, absolute volume and IT specialisation are high. HX implication: move beyond code-only use to CX/EX workflows to diffuse benefits widely.

Indonesia / Nigeria / Bolivia — emerging

Low per-capita AUI but significant upside once cloud access, skills and trust frameworks mature. HX implication: prioritise “augmentation first” public services (health triage, SME advisory) to avoid displacement dynamics.

US vs. China: Leaders with Divergent Strengths

The US and China exemplify AI’s bipolar landscape. In Anthropic data, the US boasts a 21.6% global share and AUI of 3.62, with state-level variation: District of Columbia (3.82), Utah (3.78), California (2.13), New York (1.58), Virginia (1.57). Usage diversifies beyond coding (36% global) to education (12.7%) and sciences (7.4%), with augmentation (collaborative modes) at 47%.China is absent from Anthropic’s top rankings, implying low AUI (<1, akin to emerging economies like India at 0.27). This may stem from regulatory restrictions, domestic AI alternatives (e.g., Baidu, Alibaba), or data exclusion. Inferred patterns suggest heavy coding focus (>50%, like India) and automation dominance.Stanford highlights US dominance: 40 notable models (vs. China’s 15), $109.1B investment (12x China’s $9.3B), leading highly cited publications (e.g., 669 RAI papers vs. 268). China excels in volume: 23.2% publications, 69.7% patents, 276,300 robot installations (7.3x US’s 37,600), 105 clinical trials (vs. 97). Performance gaps narrowed (e.g., MMLU: 0.3 points). Generative AI investment: US $29.04B vs. China $2.11B.WEF shows both as high-expectation economies: US 94% transformation, China 90%. China leads robot density; US in GenAI demand. Churn: US 69% AI/info processing; China high automation (45% autonomous tasks in Saudi proxy, but similar trends).

Aspect	US	China	Perspective
Usage Share (Anthropic)	21.6% (AUI 3.62)	Low (<1 AUI inferred)	US diversifies; China automation-focused
Models/Investment (Stanford)	40 models, $109.1B	15 models, $9.3B	US quality/$$ lead; China volume
Transformation (WEF)	94%	90%	Both high, China robotics edge
Publications/Patents (Stanford)	Leads cited	23.2% total, 69.7% patents	China scale advantage

These divergences risk inequality: Stanford notes US AI jobs concentrated (CA 15.7%), while WEF warns of geoeconomic fragmentation (50% in Eastern Asia). If US pulls ahead in capabilities and China in deployment scale, global gaps could widen.

Part III. Enterprise Deployment and Automation Trends

From consumer to enterprise

Anthropic’s enterprise API data reveal 77% of usage patterns are automation-dominant . Unlike consumer interactions, where augmentation and learning dominate, enterprises programmatically delegate tasks.

This matters because automation has clearer productivity effects—and sharper labour implications.

Price vs. capability

Surprisingly, enterprises show weak price sensitivity. Higher-cost tasks (measured by tokens) often dominate adoption . Capability trumps cost; firms pay more if the automation benefit outweighs the marginal fee.

PwC’s internal research on AI deployment aligns: tasks with high cognitive burden and repeatability yield the greatest ROI, even if costly to implement.

Enterprise case studies

Logistics Customs Clearance: A global freight firm deployed AI to automatically extract cargo data, fill regulatory forms, and cross-validate. Clearance times dropped from three days to a few hours.
Healthcare Insurance: A US insurer used AI to process handwritten doctors’ notes into structured claims. Accuracy exceeded 90%, with human review only on edge cases, cutting adjudication costs by 35%.
Retail Dynamic Pricing: A multinational retailer replaced brittle pricing rules with AI agents that adjusted promotions in real time based on competitor signals and supply. Pilot markets saw a 5% margin uplift.

Industries and Sectors: Task Specialisation and Shifts

Anthropic shows global tasks dominated by coding (36.9%), education (12.7%), office/admin (8.4%), arts/media (8.5%), and sciences (7.4%). Over time (Jan–Aug 2025), education rose +3.4pp, sciences +1.1pp; business/finance fell -2.8pp. API (enterprise) skews more specialized: coding 44%, office/admin 10%, education 3.6% (down from 12.3%), arts 5.2% (down from 8.2%). Automation: 77% in API vs. 49.1% in Claude.ai.Bottom-up clusters: API top uses include debugging web apps (6.1%), frontend code (6.0%), business software (5.2%), AI development (4.9%), marketing content (4.7%). Complex tasks correlate with longer inputs/outputs, suggesting context bottlenecks (elasticity 0.38: 1% input increase yields 0.38% output rise). Cost-usage positive (elasticity 2.96 overall, -0.29 after controls), implying capability/value trumps price.Stanford reinforces: AI boosts productivity 10–45% (e.g., 14.2% more customer support, 26.08% faster software). Sectors: Healthcare (223 FDA devices, up from 6 in 2015; 537 CDS trials); transportation (Waymo 150K rides/week); energy (Microsoft $1.6B nuclear). Robotics: 541K installations (down 2.2%), co-bots 10.5%. Functions: IT 48%, marketing 47%.WEF projects AI disrupting 86% of sectors, with highest in IT (99%), finance (97%), electronics (95%). Displacement: clerical (-20–40%), data entry (-24–40%); growth: AI specialists (+31–361%), data analysts (+36–85%). Human-machine: 18% tech now to 31–42% by 2030. Barriers: skills (63% in IT/finance).

Sector/Task	Anthropic Share (Claude.ai/API)	Stanford/WEF Insights
Coding/Computer/Math	36.9%/44%	26% faster tasks (Stanford); +87% AI skills (WEF)
Office/Admin	8.4%/10%	Automation dominant (Anthropic); 63% skills barrier (WEF)
Education/Library	12.7%/3.6%	Teach AI (81% teachers plan, WEF); gaps in access (Stanford)
Healthcare	7.4% (sciences)	223 FDA devices (Stanford); 92% AI trend (WEF)
Arts/Media	8.5%/5.2%	Creative thinking +61–70% (WEF)

Anthropic: Coding 36.9%/44% (Claude.ai/API), education 12.7%/3.6%, automation 49.1%/77%. Cost elasticity positive (2.96).

McKinsey: Top sectors: Tech (85% adoption, 40% cost savings), finance (78%, 35% revenue up), healthcare (72%, 28% efficiency). GenAI: 61% for content creation, 55% coding. Risks: 45% report inaccuracies.

PwC: Sectors: Manufacturing +26% productivity, services +14%. Job shifts: +5.3M net jobs 2025, but -2.1M in routine tasks.State of AI: Compute-intensive sectors (e.g., biotech AI models up 300%) lead; energy AI reduces emissions 10–15%. Benchmarks: Multimodal models +20% accuracy.

OECD: Sector readiness: Digital 90%, agriculture 40%. Policy: 70% subsidies for AI in high-value sectors.Stanford (healthcare 223 devices) and WEF (IT 99% disruption) align, with McKinsey quantifying savings.

Sector	Anthropic Share	McKinsey/PwC Insights	State of AI/OECD
Coding/Tech	36.9%/44%	85% adoption, 40% savings	+20% accuracy
Finance	3.1%	78%, 35% revenue	High subsidies
Healthcare	7.4%	72%, 28% efficiency	Biotech +300%
Manufacturing	N/A	+26% productivity	Emissions -10–15%

Part IV. Sectoral Differences in Adoption

Uneven distribution

The US Census Bureau’s Business Trends and Outlook Survey shows AI adoption rose from 3.7% of firms in 2023 to 9.7% in 2025 . Yet adoption varies dramatically:

Information sector: 25% adoption.
Accommodation & Food Services: <3%.

Stanford HAI confirms this sectoral skew: knowledge-worker-dense industries are early movers, while labour-intensive sectors lag.

Case studies

Public Sector – Estonia: AI assistants handle 70% of citizen queries across government services. The result is both higher citizen satisfaction and reduced administrative burden.
Hospitality – Japan: A Tokyo hotel chain experimented with AI concierges. While novelty attracted attention, lack of integration with hospitality systems limited efficiency gains—illustrating barriers in service-heavy industries.

Inequality and Broader PerspectivesAnthropic warns of divergence: high-AUI countries diversify tasks and favor augmentation (47%), low-AUI automate more (after controls: 3.1pp drop in automation per AUI unit, R²=0.394). If gains concentrate in rich regions, inequalities widen—reversing recent convergence.Stanford highlights gaps: gender (69.5% male talent), access (Africa 34% electrified schools; US rural CS 56% vs. urban 70%), bias (small datasets risk overfitting), emissions (GPT-4 5,184 tons CO2). Economic: investment skewed (US 12x China); data restrictions up 15–26pp, shrinking commons.WEF emphasizes risks: 92M displaced by 2030, churn 22–39% (higher in emerging economies, e.g., 48% Egypt). 41% employers downsize for AI; barriers in low-income (skills 63%, culture 46%). Yet opportunities: AI boosts trade 34–37% by 2040 if gaps bridged; digital inclusion counters inequality. Perspectives: Human-centric AI (redesign roles, WEF); global governance for trust (WEF).

Inequality Dimension	Anthropic	Stanford	WEF
Geographic	High AUI-income correlation	US concentration (CA 15.7% jobs)	Emerging churn 48%; low-income barriers
Gender/Access	N/A	69.5% male; rural gaps	Women ICT grads ~25%; skills gaps 63%
Economic/Labor	Automation in low AUI	Emissions rise; data shrink	92M displaced; 41% downsise

Hypotheses and Implications

From Anthropic: Maturity hypothesis—higher AUI diversifies beyond coding (e.g., India >50% coding vs. global 36.9%); automation declines with adoption. Inequality risk: Gains in rich regions reverse convergence.Stanford: Surge hypothesis—adoption jumps 23pp, but gaps perpetuate (e.g., gender stable since 2016). Policy: Regulations up (US 59), but incidents +56.4%.WEF: Disruption hypothesis—AI displaces but nets +78M jobs; skills shift to analytical/AI (69–88%). Inclusive growth: Reskilling (85% priority), human-centric design.Implications: Policymakers should address digital divides (WEF: broaden access), invest in context curation (Anthropic), and regulate responsibly (Stanford: 233 RAI incidents). If unaddressed, AI could exacerbate inequalities; proactive measures could boost trade 37% (WTO via WEF).

Part V. The Human Experience (HX) Dimension

As I have argued elsewhere, HX = CX + EX. The true measure of AI adoption is not just efficiency but experience equity.

Customer Experience (CX)

AI reshapes CX through speed, personalisation, and availability:

Banking: AI chat agents resolve 60% of retail banking queries instantly.
E-commerce: personalised recommendations drive 10–15% uplift in conversion.
Healthcare: triage bots reduce waiting times by 30%.

Employee Experience (EX)

Employees benefit when AI eliminates drudgery and cognitive overload. PwC pilots show lawyers using AI contract review tools reduced time spent on first-draft analysis by 35%. Far from replacing legal judgement, AI enabled lawyers to focus on negotiation and nuance.

Case Study –

HX in Insurance

One insurer restructured claims handling around AI. Customers received faster payouts (CX uplift), while employees reported higher job satisfaction as they focused on complex adjudications (EX uplift). Net Promoter Score rose 12 points in one year.

Part VI. Divergence Risks and Policy Implications

Automation vs. augmentation

The Anthropic Index highlights a paradox:

Low adoption economies rely more on automation (delegating complete tasks).
High adoption economies tend to use AI collaboratively (augmentation) .

This suggests that advanced economies may gain by complementing skilled workers, while emerging economies risk displacement of routine labour.

Labour market impacts

Brynjolfsson, Chandar, and Chen (2025) show entry-level workers with high AI exposure face worse employment prospects since 2022. Conversely, experienced workers see higher demand, consistent with AI complementing tacit knowledge.

Policy levers

Skills passports and retraining: Singapore’s SkillsFuture offers a model of continuous adaptation.
Governance frameworks: EU AI Act, UK AI Safety Institute, and ISO/IEC 42001 provide assurance mechanisms.
Infrastructure equity: Cloud credits and compute sharing can help emerging economies leapfrog.

Part VII. Looking Forward: Scenarios to 2035

The Commonwealth Scenario

In this utopian trajectory, AI adoption is democratised. Infrastructure, governance, and skills policies converge to distribute benefits. AI augments rather than replaces, creating new jobs in governance, purpose, and recognition (MIT Sloan, 2025). Global GDP growth accelerates, and inequality narrows.

The Fortress Scenario

In this dystopian arc, adoption concentrates in wealthy economies and automation-ready sectors. Entry-level labour markets hollow out, and geopolitical competition intensifies. AI becomes a source of strategic advantage, deepening divides.

My Genesis at the Fork (2025 → 2050) framework situates us at this inflection. Whether we tilt Commonwealth or Fortress will depend on choices made in the next decade—on governance, inclusion, and HX-by-design.

Conclusion: Towards Human-Centred Adoption

AI adoption is fractal: Israel and Singapore embody augmentation-first microcosms, the EU codifies assurance into law, the US races ahead but with patchy governance, and China scales through industrial policy. These contrasts reveal that growth, work, and human experience in the age of AGI will be shaped less by capability itself, and more by how nations institutionalise trust, safety, and augmentation as public goods.

The Anthropic Economic Index makes clear: AI adoption is fast but fractured. Geography, sector, and task shape patterns of use, with affluent, knowledge-worker-heavy economies racing ahead while others lag.

History shows diffusion gaps can widen inequalities. But history also shows that policy, design, and deliberate choices can bend trajectories. The internet, once concentrated, eventually became near-universal. Electrification, once urban, eventually lit rural homes.

The Anthropic Economic Index, alongside Stanford and WEF insights, underscores AI’s uneven trajectory: concentrated in high-income areas and automation tasks, with vast potential for growth and disruption. US leads in innovation, China in scale, but global gaps risk divergence. As adoption surges (78% organizations), focusing on skills, inclusion, and governance will determine whether AI fosters equitable progress or widens divides. For full datasets, explore Hugging Face (Anthropic), Stanford HAI, and WEF reports.

The question is whether AI’s gains will converge or diverge. Firms must design AI not merely for efficiency but for HX—uplifting both customer and employee experiences. Policymakers must invest in infrastructure, skills, and governance to avoid an AI divide.

The stakes are profound: not just productivity and GDP, but fairness, dignity, and human flourishing in the age of agentic AI.

As we stand at this frontier, the imperative is clear: trust by design, equity by intent, and human experience at the centre.

References

Anthropic. Economic Index 2025: Uneven Geographic and Enterprise AI Adoption.
Stanford HAI. AI Index 2025.
PwC. AI Jobs Barometer 2025.
World Economic Forum. Future of Jobs Report 2025.
McKinsey Global Institute. The State of AI in 2025.
Gallup. AI Use at Work Has Nearly Doubled in Two Years (2025).
US Census Bureau. Business Trends and Outlook Survey (2025).
Brynjolfsson, E., Chandar, R., Chen, C. Canaries in the Coal Mine? Employment Effects of AI (2025).
MIT Sloan. Superhuman Workflows and the Rise of Agentic AI (2025).