Chapter 15 — Labour, productivity, and economic effects
This chapter develops the labour, productivity, and economic-effects framework for AI deployment. The framework integrates several research traditions (task-based labour economics; productivity measurement; technology-and-skill-premium dynamics; institutional-and-policy responses) to address the contemporary period’s most-contested question: what is AI doing to the economy and to workers?
The empirical evidence through 2024–2026 has moved substantially beyond the early-2023 macroeconomic projections that dominated public discussion. The Goldman Sachs March 2023 estimate of 300 million jobs at risk globally was widely cited and substantially shaped subsequent debate; the actual labour-market trajectory through 2024–2026 has been substantially milder. Professional-services employment grew rather than contracted (Section 11.12); manufacturing employment in advanced economies stabilised; the most-significant labour displacement has been concentrated in specific narrow contexts (junior creative roles; some routine customer-service roles; specific automated-vehicle adjacent roles) rather than spread broadly across the economy. The pattern is consistent with the Acemoglu-Restrepo (2020) task-based framework: AI substitutes for specific tasks rather than entire occupations; the labour-market effects depend on whether freed-up capacity is deployed for higher-value work (augmentation) or whether headcount reductions follow (displacement).
The chapter develops the framework with explicit reference to Part II case material. The Hollywood strike provisions (Section 10.7) are the most-detailed organised-labour response. The professional-services augmentation pattern (Section 11.12) is the contemporary white-collar test case. The Klarna customer-service trajectory (Section 8.4) is the cautionary case for premature substitution. The agricultural-labour and mining-autonomy patterns (Sections 11.7 and 9.11) provide sectoral data on more-direct displacement. The cumulative case material supports a more-nuanced analysis than the macroeconomic projections alone provided.
The chapter develops fourteen sections. Section 15.1 frames the labour-AI question. Section 15.2 develops the Acemoglu-Restrepo framework. Section 15.3 covers the 2023–2024 projections and the subsequent reality. Section 15.4 develops the augmentation-vs-displacement empirics. Section 15.5 covers sector-specific dynamics. Section 15.6 addresses the skill premium and wage effects. Section 15.7 covers productivity questions. Section 15.8 covers geographic and demographic concentration. Section 15.9 covers the skills-transition challenge. Section 15.10 covers policy responses. Section 15.11 covers the new-occupation question. Section 15.12 covers Australian and Malaysian labour markets. Section 15.13 covers institutional responses. Section 15.14 sketches the 2026–2030 forward trajectory.
15.1 The labour-AI question — framing
The labour and economic effects of AI deployment have generated more public discussion than any other dimension of the contemporary AI period. The discussion has often been polarised — between projections of mass unemployment and projections of business-as-usual — without substantial engagement with the empirical evidence accumulating through 2023–2026. The framework this chapter develops aims to support more-careful analysis grounded in the available evidence.
Why this matters. Three concerns motivate the labour-AI question.
First, employment is the primary mechanism through which most workers in market economies access income. Substantial employment displacement, if it occurred without offsetting employment growth, would produce direct welfare consequences for affected workers. The historical pattern of major technological transitions has been that aggregate employment recovers (often with substantial growth) while specific affected workers may face prolonged adjustment costs.
Second, the distribution of effects matters as much as aggregate effects. Even if AI deployment produces aggregate labour-market neutrality or growth, concentrated displacement in specific roles, regions, or demographic groups produces substantial welfare consequences for those affected. The political consequences of distributional effects can be substantial; the broader social cohesion can be affected by uneven labour-market outcomes.
Third, productivity dynamics matter for long-run economic growth. If AI deployment produces substantial productivity gains, the broader economic capacity expands; if AI deployment produces minimal productivity gains, the broader economic effect is limited. The question of how much productivity AI is actually producing is contested; the answer affects long-run economic outcomes substantially.
The historical context — prior automation waves. AI is not the first major automation wave; understanding the historical pattern provides context for contemporary analysis.
The 19th-century textile mechanisation displaced large numbers of skilled handloom weavers; the broader labour market absorbed the displacement over decades, with substantial transition costs concentrated on directly-affected workers. Aggregate employment grew substantially over the broader period.
The early-20th-century manufacturing-automation wave (mass production; assembly-line work; agricultural mechanisation) displaced agricultural labour at substantial scale; rural-to-urban migration in the developed world was driven partly by this displacement. The transition occurred over generations rather than years; the cumulative welfare effects were substantial.
The mid-20th-century computerisation wave (mainframe computers; office automation; bank-and-insurance computerisation) displaced specific clerical roles while creating new technical roles. The computerisation wave produced the skill-biased technical change pattern that subsequent labour-economic literature extensively studied.
The late-20th-century globalisation-and-automation wave (manufacturing offshoring; further computerisation; the emergence of advanced robotics) produced the contemporary structural-economic patterns that the contemporary AI moment is overlaid on. The Autor-Levy-Murnane (2003) task-based framework and the subsequent extensions (including Acemoglu-Restrepo 2020) emerged from analysis of this wave.
The contemporary AI wave shares some features with prior waves and differs in others. The shared features include: capability that can substitute for specific tasks; uneven distribution of effects across roles and geographies; substantial transition costs for affected workers. The distinct features include: the breadth of tasks that AI can address (covering substantially more of the cognitive work that prior waves did not address); the speed of capability advancement (foundation-model capability progress over months rather than years); the global-and-instant deployment patterns (cloud-based AI is deployed simultaneously across geographies in ways that prior automation was not).
The contemporary distinctness. A specific feature of the contemporary AI moment that distinguishes it from prior waves is its address to cognitive work. Prior automation waves substituted for physical and routine tasks; the contemporary AI wave substitutes for tasks involving substantial cognitive content (writing; analysis; design; certain creative work). The categories of work affected are therefore broader than prior waves; the affected workers are concentrated in different demographic and geographic patterns than prior waves affected.
The cognitive-work focus has specific implications. First, the affected workers are typically higher-income than prior automation waves’ affected workers; the welfare consequences for individual affected workers may be smaller in absolute terms but the political and social consequences may differ. Second, the cognitive content of the substitution is variable across foundation-model capability — current models substitute well for some cognitive tasks (specific routine document production; some specific analytical tasks) and poorly for others (complex judgment; novel problem-solving). The substitution pattern is therefore heterogeneous across cognitive work in ways that prior physical-and-routine-work substitution was less so.
15.2 The Acemoglu-Restrepo framework
The contemporary analytical framework for understanding labour-AI dynamics is substantially built on the work of Daron Acemoglu and Pascual Restrepo. Their 2018, 2019, 2020, and 2022 papers establish the task-based framework that subsequent literature has built on. Acemoglu’s 2024 NBER working paper applies the framework specifically to AI.
Tasks vs occupations. The foundational distinction in the framework is between tasks and occupations. Occupations are clusters of tasks that workers perform; jobs are bundled-task arrangements that produce labour-market outcomes. Automation typically substitutes for specific tasks rather than entire occupations; the labour-market effects on occupations depend on the proportion of an occupation’s tasks that are automated.
The framework’s analytical power comes from this distinction. Frey and Osborne’s influential 2017 paper estimated that 47% of US occupations were at “high risk of computerization” by treating occupations as wholes; the Acemoglu-Restrepo framework decomposes the analysis to the task level and produces substantially different (and more-nuanced) estimates. An occupation with 70% of tasks automatable does not necessarily disappear; it may transform with workers focusing on the 30% non-automated tasks while completing the 70% with AI augmentation.
Substitution, augmentation, and reinstatement. The framework identifies three distinct effects of automation on labour demand.
Substitution (sometimes called displacement) occurs when automation directly replaces human labour in performing specific tasks. The displaced workers must reallocate to other tasks within the same occupation, to different occupations, or to non-employment. The substitution effect depresses labour demand.
Augmentation occurs when automation makes human workers more productive at remaining tasks. Augmentation increases labour demand for the affected workers (more productive workers are valuable; firms employ more of them) but may decrease labour demand for less-augmented workers (substitution within the labour market away from less-productive workers).
Reinstatement (sometimes called task creation) occurs when automation creates new tasks that did not previously exist. The new tasks may be performed by humans (creating new labour demand) or by other automation. Historical examples include programming jobs created by computer development; software-engineering jobs created by software development; AI-engineering jobs created by AI development.
The net labour-market effect of automation depends on the balance among these three effects. Acemoglu and Restrepo’s empirical work on industrial robots (2020 Journal of Political Economy) found that robot adoption in US labour markets produced substantial substitution effects with limited reinstatement, leading to net employment-and-wage declines in directly-affected commuting zones. The pattern is informative for contemporary AI analysis but does not necessarily generalise — AI may have different substitution-augmentation-reinstatement balances than industrial robots.
Empirical methodology. The Acemoglu-Restrepo methodology combines sectoral and regional data with task-level analysis to identify the different effects. The methodology has been substantially adopted in subsequent literature; specific applications to AI through 2023–2026 have used variants of the approach. The methodology is not without limitations: distinguishing the three effects in observational data is difficult; the effects may be heterogeneous across contexts in ways the methodology partially captures; the time-lags between technology adoption and labour-market effects are substantial and complicate identification.
Acemoglu’s 2024 AI-specific analysis. Daron Acemoglu’s NBER Working Paper “The Simple Macroeconomics of AI” (April 2024, NBER WP 32487) applied the framework specifically to AI deployment through 2024. The analysis estimated: the share of US labour-market tasks that AI can profitably automate at approximately 4.6% over the next decade; the productivity gain from this automation at approximately 0.55–0.71%; the GDP impact at approximately 0.93–1.16% over the next decade. The estimates are substantially smaller than the Goldman Sachs and McKinsey projections (Section 15.3) and align more closely with the actual 2024–2026 empirical pattern. The paper has been influential as a counterweight to the more-aggressive macroeconomic projections; the methodology and assumptions are contested but the framework approach is broadly accepted.
15.3 The 2023–2024 macroeconomic projections vs reality
The contemporary public discussion of AI’s labour-market effects was substantially shaped by a series of 2023 macroeconomic projections that have not been borne out in subsequent data.
The Goldman Sachs March 2023 projection. Goldman Sachs Global Investment Research published The Potentially Large Effects of Artificial Intelligence on Economic Growth in March 2023. The analysis estimated: approximately 300 million jobs globally could be exposed to automation by AI; approximately 18% of global work could be automated by AI; productivity growth could be substantially boosted (potentially 1.5 percentage points annually). The 300 million figure dominated public discussion; the projection was widely cited in subsequent commentary.
The methodology aggregated occupational task profiles, applied estimates of which tasks AI could perform, and computed exposure totals. The methodology had specific limitations: it assumed that exposed tasks would be automated rather than augmented; it did not account for the deployment-environment friction that subsequent chapters of this textbook have developed; it did not adequately account for productivity-and-growth effects that produce reinstatement-style new task creation.
The Brynjolfsson-Li-Raymond 2023 customer-service study. A specific influential study was Brynjolfsson, Li, and Raymond’s 2023 NBER working paper Generative AI at Work (NBER WP 31161). The study examined the introduction of a generative AI assistant to approximately 5,000 customer-service agents at a Fortune 500 firm; the AI tool produced approximately 14% productivity gain on average, with the gain concentrated among less-experienced workers (35% gain for the bottom-quartile-skill workers; minimal gain for top-quartile workers). The study was widely cited as evidence that AI augmentation produced substantial productivity benefits with particular value for less-skilled workers.
The study was significant for the methodology (a controlled rollout supporting causal inference) and for the empirical findings (the augmentation pattern with skill-equalisation effects). Subsequent attempts to replicate or extend the findings produced more-mixed results; the contemporary consensus is that the Brynjolfsson-Li-Raymond effects are real but may not generalise across all contexts. The Klarna case (Section 8.4) is structurally relevant: the Klarna deployment was customer-service AI that was meant to produce similar augmentation but was deployed in a substitution mode that produced worse outcomes than the augmentation-focused Brynjolfsson-Li-Raymond context.
Subsequent revisions. Through 2024–2025, multiple updated projections have produced substantially smaller estimates than the original 2023 projections. The McKinsey Global Institute’s 2024 Generative AI in the Workplace report estimated potential automation at 30% of work-hours in advanced economies by 2030 — substantially lower than the original 2023 estimates and with more-nuanced treatment of the augmentation-vs-substitution balance. The IMF’s 2024 Gen-AI: Artificial Intelligence and the Future of Work analysis similarly estimated more-modest aggregate effects. The Acemoglu 2024 estimates (Section 15.2) at 4.6% of tasks and approximately 1% GDP impact are at the conservative end of the post-revision range but are increasingly cited in academic and policy discussion.
The actual labour-market trajectory through 2024–2026. The empirical labour-market data through 2024–2026 has not shown the substantial employment displacement that the original projections suggested. US unemployment remained low through 2024 (3.7–4.2% range); US payrolls grew substantially; major occupational categories that the projections had identified as exposed (legal services; accounting; software engineering; customer service; creative work) showed continued employment growth or stable employment rather than contraction.
The pattern requires careful interpretation. Several factors may be operating:
- AI deployment may not yet have reached the scale that produces substantial labour displacement. Foundation-model capability is improving, but the share of work that is actually being substituted by AI in production deployment in 2026 is small.
- Augmentation effects may be dominating substitution effects. The Brynjolfsson-Li-Raymond pattern (productivity-with-employment-stable) may be the more-common deployment pattern than substitution.
- Reinstatement effects may be substantial. New AI-related occupations (prompt engineering; AI safety; AI compliance; AI ops) plus expanded employment in firms that benefit from AI-driven efficiency may be offsetting substitution effects.
- The time-to-impact may be longer than projections assumed. Labour-market effects of major technologies typically lag deployment by years; the 2024–2026 lack of major effects does not preclude future effects.
The empirical pattern through 2026 supports the Acemoglu 2024 estimates substantially more than the 2023 Goldman Sachs projections. The macroeconomic AI effects appear to be modest and accumulating rather than transformative and immediate.
15.4 The augmentation-vs-displacement empirics
The augmentation-vs-displacement balance is the central empirical question for AI labour-market analysis. The contemporary evidence supports specific conclusions about where each effect dominates.
Where augmentation dominates. Several contexts show clear augmentation patterns:
Software engineering — GitHub Copilot, Cursor, and adjacent tools are augmenting developers rather than replacing them. The Microsoft-GitHub research (Peng et al., 2023) showed substantial productivity gains for developers using Copilot; subsequent research has generally confirmed productivity gains while not finding substantial employment displacement. The 2024 software-engineering job market has been more competitive than the 2021–2022 boom, but this reflects broader tech-industry restructuring rather than AI-driven displacement.
Professional services (Section 11.12) — Legal, accounting, and consulting employment has grown rather than contracted through 2024–2026. The augmentation pattern (described in detail in Section 11.12) has been the dominant deployment mode.
Healthcare (Chapter 7) — Clinical AI has been deployed primarily as augmentation of physician work rather than substitution. Specific applications (radiology AI; ambient scribes; clinical-decision-support systems) augment specific tasks; physician employment has continued to grow.
Customer service in well-managed deployments — The Brynjolfsson-Li-Raymond pattern of augmentation with productivity gains has been demonstrated in multiple contexts where the deployment is structured for augmentation rather than substitution.
Where displacement effects appear. Several contexts show more-direct displacement:
Junior creative roles in advertising and design — Some junior copywriting, design, and content-production roles have been substituted by AI. The displacement is concentrated at the entry level; senior creative work continues to be performed by humans (Section 10.2).
Content moderation — AI-driven content moderation has displaced some human moderation roles, though the broader content-moderation work continues to require human judgment for complex cases. The Trust and Safety industry employment has been pressured but not collapsed.
Specific routine clerical and customer-service roles — Certain routine inquiry-handling roles, particularly at firms with aggressive substitution-focused AI deployments, have seen substantial reduction. The Klarna case (Section 8.4) is the public reference; the actual pattern across the industry is varied.
Some translation work — Translation services, particularly for routine documents, have substantially shifted toward AI-augmented or AI-direct production. Specialised translation (literary; legal; technical) continues to require human work.
Specific autonomous-vehicle adjacent work — Some long-haul trucking, taxi-driving, and adjacent work is exposed; the actual displacement through 2026 has been limited because autonomous-vehicle deployment has been slower than projected (Section 12.4).
Where the balance is unclear. Several contexts have ambiguous patterns:
Knowledge-work middle layers — Mid-career analyst, associate, and adjacent roles in professional services and similar industries have ambiguous trajectories. Senior professionals are increasingly augmented; junior professionals’ work is increasingly automatable; the mid-career role’s continued necessity is unclear.
Education — Teaching has not been substantially displaced through 2026, but AI-tutoring deployment continues to grow. The role-of-teacher question (Section 12.2) is unresolved.
Creative work generally — The Hollywood strike provisions (Section 10.7) have produced specific protective frameworks for organised creative labour, but unorganised creative work (independent designers; writers; musicians) has more-uncertain trajectories.
The unevenness across roles within occupations. A specific empirical pattern is the unevenness within occupations. Within software engineering, junior implementation-focused roles are more substitutable than architecture-focused roles. Within legal practice, document-review-focused roles are more substitutable than client-relationship-focused roles. The pattern is consistent with the task-based framework: the unevenness reflects which specific tasks are most-automatable.
The McKinsey, BCG, and EY analyses. The major consulting firms have produced extensive analysis through 2023–2026. The McKinsey Global Institute’s Generative AI in the Workplace report (May 2024) identified specific occupations and tasks at greatest exposure; the BCG analysis through 2024–2025 produced similar work; EY’s parallel analyses extended the methodology. The analyses converge on specific themes: the deployment pattern is more augmentation than displacement; the affected roles are specific rather than universal; the productivity gains are real but more modest than 2023 projections suggested; the pace of change is gradual rather than abrupt. The analyses are not without their own biases — the consulting firms benefit commercially from AI deployment activity — but the convergent themes from independent analytical efforts are informative.
15.5 Sector-specific labour dynamics — the Part II evidence
The Part II case material provides specific sector-by-sector evidence on labour-AI dynamics. The patterns differ across sectors in informative ways.
Hollywood and creative industries — the organised-labour response. Section 10.7 covered the 2023 WGA and SAG-AFTRA strikes in detail. The contract provisions secured by the unions are the most-detailed organised-labour AI response in any major industry: AI-generated material cannot serve as source material; writers retain residuals on AI-augmented work; performers must consent to digital replicas with specific compensation structures. The provisions substantially constrain AI deployment in Hollywood production; the deployment patterns reflect this constraint.
The structural lesson is that organised labour can substantially shape AI deployment trajectory in industries where labour is organised. The lesson generalises but is not universal: industries without comparable organisation have not produced comparable protective frameworks.
Professional services — the augmentation pattern. Section 11.12 covered the augmentation pattern in detail. The augmentation-with-shifted-task-mix dynamic has been the dominant pattern; aggregate professional-services employment has grown; the specific role-mix has shifted toward more judgment-and-relationship work and less routine production.
The structural lesson is that high-skilled-labour augmentation, when structured appropriately, can produce productivity gains without substantial displacement. The lesson is conditional on the structuring; the Klarna pattern (substitution rather than augmentation) demonstrates that the same technology in the same broad sector can produce different labour outcomes depending on deployment strategy.
Customer service — the cautionary case. Section 8.4 covered the Klarna deployment in detail. The substitution-focused deployment produced documented operational problems and substantial reputational cost; the subsequent reversal involved rehiring substantial numbers of human customer-service agents. The case is the contemporary reference for what premature substitution-focused deployment looks like.
The structural lesson is that customer-service AI deployment is technically feasible but operationally challenging when structured for direct substitution. The augmentation pattern (Brynjolfsson-Li-Raymond and similar) is more reliable than the substitution pattern. The 2024–2026 industry consensus has shifted toward augmentation; the post-Klarna deployment pattern is substantially more cautious than the early-2024 trajectory.
Manufacturing — the autonomy pattern. Section 9.4 covered manufacturing autonomy. The pattern is heterogeneous: some specific applications (Foxconn lights-out manufacturing; Tesla’s robot-augmented production) have substituted for specific roles, but aggregate manufacturing employment in advanced economies has been stable rather than contracting through 2024–2026. The heterogeneity is partly because manufacturing has been substantially automated for decades — the AI wave is incremental rather than revolutionary in this sector.
Mining and resources — the displacement frontier. Section 9.11 covered the Australian mining-autonomy frontier. The Pilbara remote-operations centres have substituted for specific on-site roles; the displaced workers have largely been redeployed to operations-centre roles or have transitioned to other employment. The cumulative employment impact has been notable but not catastrophic; the transition has been managed over a decade-plus timeframe.
Agriculture — the smallholder problem. Section 11.7 covered the agricultural labour question. Agricultural-labour displacement is concentrated in advanced-economy large-farm contexts where autonomous equipment is deployed; the broader smallholder agriculture (which dominates global agricultural employment) is largely unaffected because the deployment economics do not support smallholder adoption. The pattern is structurally distinct from other sectors: rather than displacement-vs-augmentation within the same economy, the agricultural pattern is displacement in advanced economies and minimal effect in developing economies.
The cumulative pattern. The sectoral evidence supports the broader analytical framework: AI labour effects are heterogeneous; the augmentation-displacement balance varies across sectors; the broader employment effects are modest aggregate-level but substantial for specific role categories. The pattern is consistent with the Acemoglu-Restrepo task-based framework rather than with the more-aggressive 2023 macroeconomic projections.
15.7 The productivity question
Beyond labour-market effects, the productivity question is what AI does to overall economic capacity. The question is contested; the empirical evidence is still accumulating.
The productivity paradox. A specific concern through 2023–2026 has been the productivity paradox — the apparent disconnect between substantial AI investment and modest measured productivity growth. US labour-productivity growth through 2023–2024 has been moderate (1–2% annually), substantially below the projections that aggressive AI deployment would produce.
The productivity paradox has historical precedent. Solow’s 1987 observation (“You can see the computer age everywhere but in the productivity statistics”) preceded a substantial productivity acceleration in the late 1990s and early 2000s; the pattern has been described as J-curve productivity dynamics where new technologies produce immediate organisational disruption (suppressing productivity in the short term) before producing substantial productivity gains as organisations adapt.
The Brynjolfsson-Hitt time-to-impact framework addresses this directly. Their work on prior IT investments found that productivity gains lag deployment by years (typically 5–7 years), as organisations adjust complementary investments (training; process redesign; organisational structure) to support the technology. Applying the framework to AI suggests that substantial productivity effects from 2023–2024 deployment may not be visible in macroeconomic data until 2028–2030.
The Acemoglu 2024 analysis. The Acemoglu 2024 NBER WP estimates productivity gains from AI at 0.55–0.71% over a decade — substantially more modest than the J-curve projections would suggest. The methodology is bottom-up: estimating the share of tasks that AI can automate, multiplied by the productivity gain on each automated task, summed across the economy. The methodology is sensitive to assumptions about which tasks AI can productively automate and what the per-task productivity gain looks like.
The Acemoglu estimates have been contested. Critics argue that the methodology underestimates AI’s potential by focusing on directly-automatable tasks and not adequately accounting for new task creation, broader cognitive complementarity, and the longer-term capability advancement. Defenders argue that the methodology is appropriately bottom-up and that more-aggressive estimates lack empirical grounding. The contest is informative; the productivity question is unsettled in 2026.
The micro-vs-macro evidence. A specific tension in the productivity literature is between micro-level evidence (specific firms and contexts showing substantial productivity gains) and macro-level evidence (aggregate productivity data showing modest gains). Both can be true simultaneously: AI may be producing substantial productivity gains at the firm level for specific firms and applications, but the aggregate effect remains modest because adoption is uneven and the overall economy is not yet benefiting at scale.
The micro evidence includes: Brynjolfsson-Li-Raymond customer service (14% gain); various coding-productivity studies (10–55% gains depending on context); specific case studies of AI-driven cost reductions at major operators. The macro evidence includes: relatively stable aggregate productivity growth; uneven deployment across firms and sectors; substantial productivity dispersion across firms within sectors.
The reconciliation is consistent with the J-curve dynamics: the micro effects are accumulating but the macro effects lag. The 2026–2030 trajectory will produce more-substantial macro evidence; the question is whether the pattern matches the late-1990s computer-driven productivity acceleration or whether AI produces a different (perhaps more-modest) trajectory.
The methodology question. A specific challenge for AI productivity measurement is methodological. Standard productivity measures (GDP per hour worked; total factor productivity) are designed for industrial-economy production; their application to knowledge-economy and digital-economy work is contested. Some commentators argue that AI’s productivity contribution is systematically underestimated by standard measures because the value of digital services is mismeasured. Others argue that standard measures are adequate and that the modest measured productivity gains accurately reflect modest actual gains. The methodology question itself is unsettled.
15.8 Geographic and demographic concentration
A specific dimension of labour-AI effects is the concentration of effects in specific geographies and demographic groups. The concentration affects the welfare implications of aggregate effects substantially.
The geographic dimension. AI labour effects vary substantially across geographies for specific reasons.
Within advanced economies, the effects are concentrated in technology-deploying regions (San Francisco Bay Area; New York; Boston; Seattle; London; specific UK cities; major Australian cities; specific Canadian cities) where AI development and deployment is concentrated. Other regions of the same economies are less directly affected.
Across advanced and developing economies, AI deployment is heavier in advanced economies, producing direct productivity-and-displacement effects there; developing economies are less directly affected by AI deployment but may face indirect effects through trade-and-investment patterns. The global geographic concentration produces specific effects: business-process-outsourcing locations (India, Philippines, Vietnam) face displacement risk if AI substitutes for outsourced work; manufacturing-export economies face complex dynamics depending on whether automation reduces the labour-cost advantage that drove their position.
Within developing economies, the pattern is mixed. Some specific applications (agricultural smallholder applications; basic financial services through mobile platforms) provide direct AI-augmented productivity gains. Other applications (advanced AI for specific high-value tasks) are largely deployed in advanced-economy operations; the developing-economy participation is via the providing of cheap-labour-substituting AI services rather than as users of advanced AI capability.
The demographic dimension. AI labour effects vary substantially across demographic groups.
Age — Older workers in displaced occupations face larger transition costs than younger workers; reskilling is harder in mid-and-late-career; pension-and-employment-history considerations complicate transition. Younger workers face the entry-level-displacement issue: AI substitutes most for the routine work that has historically been the entry path into many occupations. The career-progression problem (Section 11.12) generalises across many sectors.
Gender — Specific sectors with high female employment shares (administrative; customer service; certain healthcare roles; teaching) face different exposure than sectors with lower female employment shares. The aggregate gender-effect of AI deployment is contested; specific studies have produced mixed results.
Race and ethnicity — Where AI deployment systematically affects industries with specific demographic composition, the differential impact is substantial. The Robodebt case (Section 12.1) is structurally relevant: the affected populations were disproportionately economically vulnerable; the harms were demographically concentrated.
Education — The skill-equalisation-vs-skill-amplification question (Section 15.6) plays out across education levels. The empirical evidence is mixed; specific contexts show different patterns.
The cross-country dynamics. Different countries face different labour-AI dynamics depending on their economic structure. The United States has the most-deployed AI in absolute terms but also the largest absorptive capacity. China has substantial domestic AI deployment in a controlled regulatory environment. The EU has substantial deployment but with the AI Act’s protective framework. Smaller economies (Australia; Singapore; Malaysia; the Nordic economies) have specific deployment patterns shaped by their specific industrial structures.
The cross-country differences have implications for international policy coordination and for individual countries’ policy responses. No single policy template fits all countries; the adaptive policy responses are themselves an active area of development.
15.9 The skills-transition challenge
The skills-transition question — what skills workers need to thrive in an AI-augmented economy, and how those skills are developed — is central to labour-policy responses.
What skills are needed. Several skill categories appear durably valuable in AI-augmented contexts:
Specific technical skills for AI deployment — prompt engineering, AI integration, AI governance and compliance, AI safety. These skills are directly created by AI deployment and have substantial demand growth.
Higher-order cognitive skills — complex problem-solving, strategic thinking, judgment under uncertainty, ethical reasoning. These skills complement AI capability rather than competing with it.
Interpersonal skills — relationship management, team leadership, customer relationship building, negotiation. These skills are largely outside foundation-model capability and remain valuable.
Domain expertise — deep knowledge of specific industries, regulations, customer needs, technical contexts. Domain expertise typically combines with AI augmentation to produce expert-AI-augmented work that is more valuable than either component alone.
Adaptability and continuous learning — the ability to adapt to changing technology, learn new tools, and maintain relevance over time. Given the pace of AI capability advancement, this meta-skill is itself substantially valuable.
Reskilling and upskilling. The reskilling challenge — helping displaced workers transition to new occupations — is substantial. Historical reskilling efforts have produced mixed outcomes; the empirical literature on retraining program effectiveness is generally pessimistic about short-term retraining for major occupational transitions.
Specific reskilling-program features that improve effectiveness include: integration with local labour-market needs; substantial duration (months rather than weeks); industry partnerships ensuring relevant content; income support during retraining; placement assistance after completion. The 2024–2026 reskilling initiatives have produced variable outcomes; specific programs (the German Kurzarbeit + retraining model; the Singapore SkillsFuture program; specific US workforce-development initiatives) have produced documented success in specific contexts.
The structural-vs-cyclical question. A specific question is whether AI-related labour displacement is structural (permanent shift in labour-market composition requiring substantial reskilling) or cyclical (temporary disruption that the economy will absorb without major intervention). The historical pattern of major technological transitions has been substantially structural over long periods (decades) but with substantial cyclical variation during shorter periods. The contemporary AI moment likely fits this historical pattern; the policy implications are that both structural responses (long-term skills development) and cyclical supports (transitional income; placement assistance) are appropriate.
The retraining-effectiveness evidence. The empirical literature on retraining effectiveness is substantial and generally cautious. Card, Kluve, and Weber (2018) review of active labour-market policies found generally modest effects of training programs in producing employment outcomes. Specific programs with strong design (employer partnerships; industry-relevant skills; substantial duration) produce stronger effects. The implication: retraining is part of the policy response but not a sufficient response on its own; broader policy frameworks (labour-market support; income support; broader education investment) are needed alongside.
15.10 Policy responses — UBI, portable benefits, reskilling
The policy landscape for labour-AI dynamics is developing through 2024–2030. Several policy responses are being debated or implemented.
The Universal Basic Income debate. Universal Basic Income — a guaranteed minimum income for all residents regardless of employment status — has been substantially discussed as a response to AI-driven labour displacement. The idea has prominent advocates (Andrew Yang in the US; various European political figures; Sam Altman has contributed funding to UBI-related research) and substantial critics.
The empirical evidence on UBI effects is limited. Specific pilot programs (the OpenResearch UBI pilot; specific Finnish and Canadian pilots; various US city pilots) have produced modest evidence on small-scale effects. The broader question of whether UBI at scale would produce intended welfare gains without significant labour-market disincentive effects is unresolved; political feasibility of universal-basic-income at meaningful levels in major economies has been low through 2024–2026.
The 2024–2026 policy reality is that UBI remains a discussion topic rather than an implemented policy in major economies. The discussion has informed adjacent policy debates (negative income tax; expanded unemployment insurance; specific income-support programs) without producing major UBI implementation.
Portable benefits. A more-incremental policy response is portable benefits — health insurance, retirement savings, disability protection, paid leave — that are not tied to specific employment relationships. The framework supports more-flexible labour-market structures (gig work; contracting; entrepreneurship) without sacrificing the worker-protection benefits that traditional employment provides.
Portable benefits have been progressively implemented in specific contexts. California’s AB-2257 (codifying gig-worker classifications), state-level portable-benefits laws in several US states, and specific company-level initiatives have produced incremental progress. The 2024–2026 policy trajectory has been gradual; the broader implementation faces substantial political and operational challenges.
Reskilling programs. Public-sector reskilling programs have been a substantial policy response. The Workforce Innovation and Opportunity Act (US, 2014, with subsequent modifications), the European Social Fund Plus framework, the Singapore SkillsFuture program, and various national programs constitute the contemporary reskilling infrastructure. The 2024–2026 expansions have included specific AI-related programs at multiple levels.
The reskilling effort is large in absolute terms but small relative to the potential displacement scale. The 2026 US Workforce Innovation and Opportunity Act funding is approximately USD 3 billion annually; the comparable EU program is similar in magnitude. The funding is a small fraction of the cumulative potential displacement effect. The mismatch suggests that current reskilling efforts are inadequate if the displacement effects materialise at the scale that aggressive projections suggest.
Labour-market policy frontier. Beyond UBI, portable benefits, and reskilling, the labour-market policy frontier includes: minimum-wage increases (linked partly to AI-driven productivity gains); collective-bargaining frameworks (to support negotiation of AI deployment terms); regulatory limits on AI deployment (the EU AI Act’s high-risk-employment provisions; NYC Local Law 144’s audit requirements); tax policies affecting automation incentives (some discussion of “robot taxes”; differential corporate-tax treatment for AI-vs-labour expenses); industry-specific labour-protective frameworks (the Hollywood strike provisions; analogous provisions in other industries).
The cumulative policy framework is incremental; no single jurisdiction has produced comprehensive labour-AI policy. The 2026–2030 trajectory will produce continued incremental development; the broader policy adaptation to AI labour effects will be substantial but uneven.
15.11 The new-occupation question
A specific dimension of labour-AI dynamics is the question of what new occupations emerge from AI deployment. The reinstatement effect (Section 15.2) operates partly through new occupation creation; the historical pattern of major technological transitions has been that substantial new occupations emerge over time.
What new occupations have emerged. Through 2023–2026, several new occupational categories have emerged or expanded substantially:
Prompt engineering — Specialised work designing prompts for foundation models. The category emerged in 2022–2023 with substantial salary premiums; by 2025–2026 the category has substantially absorbed into broader software-engineering and product-management roles, suggesting it was a transitional rather than permanent specialisation.
AI safety, alignment, and policy — Substantial growth in specialised work on AI safety research, policy, governance, and adjacent areas. The category includes both technical roles (interpretability research; red-teaming; capability evaluation) and policy roles (compliance; governance; ethics). Major AI firms (Anthropic, OpenAI, DeepMind, Microsoft, others), AI safety research organisations (Apollo Research, METR, ARC, AISI/AISI-equivalents), academic centres, and policy organisations have substantially expanded employment in this category.
AI compliance and governance — Section 14.13 covered the compliance landscape; the resulting employment growth has been substantial. AI governance professionals; AI compliance officers; AI auditors; AI ethics officers; specific compliance roles in firms deploying AI. The category is growing rapidly through 2024–2026.
AI-augmented professional services — Specific roles that combine traditional professional expertise with AI tooling expertise. Lawyers specialising in AI deployment; accountants specialising in AI audits; consultants focusing on AI strategy; medical professionals specialising in AI-augmented practice. The roles are not entirely new but represent specialisations that the AI wave has produced demand for.
AI infrastructure and operations — MLOps (machine-learning operations) engineers; foundation-model operations specialists; data-engineering specialists working with foundation-model deployment. The category has substantial growth.
AI product and design — Product managers and designers specialising in AI-product development. The category combines traditional product-management with AI-specific considerations.
The historical pattern. The historical pattern of major technological transitions is informative. The internet wave (1995–2010) created substantial new occupations: web developers; SEO specialists; social-media managers; e-commerce specialists; and many adjacent categories. The smartphone wave (2008–2020) created mobile developers; app marketers; and many adjacent roles. The pattern of major technological transitions has been substantial new occupation creation over decade-plus periods.
The contemporary AI wave appears to be following a similar pattern. The new occupations that have emerged through 2023–2026 are substantial; the long-run trajectory likely produces continued new-occupation creation through 2030 and beyond.
The contemporary evidence on net effects. A specific question is whether new occupation creation is sufficient to offset displacement. The historical pattern has produced net employment growth over long periods; the question for AI is whether the same pattern holds.
The empirical evidence through 2024–2026 has been broadly consistent with offsetting effects: aggregate employment has grown rather than contracted; the specific new occupations are substantial in scale; the displaced workers have substantially absorbed into other employment. The pattern may not hold indefinitely (specific demographic and geographic effects may not absorb as easily; the future-AI-capability trajectory may produce more substantial displacement than current capability does), but through 2026 the pattern has not been catastrophic.
The long-run trajectory. The long-run trajectory question — whether AI eventually produces substantial structural unemployment, or whether the historical labour-market adaptation pattern continues — is one of the contemporary period’s most-contested questions. The answer depends on factors that are themselves uncertain: how rapidly AI capability advances; whether AI substitutes for tasks that the labour market has historically absorbed displaced workers into; whether new occupations continue to emerge at scale. The 2026–2030 evidence will substantially shape the assessment; the answer is unlikely to be definitively resolved before then.
15.12 Australian and Malaysian labour markets
The Australian and Malaysian labour markets have specific characteristics that shape labour-AI dynamics differently from US or EU contexts.
Australian labour-market dynamics. Australia’s labour market through 2024–2026 has been tight: unemployment in the 3.5–4.5% range; substantial wage growth in specific sectors; persistent labour shortages in healthcare, construction, agriculture, and several other categories. The tight labour market has supported wage growth across the income distribution and has made AI-driven displacement less acute than projections suggested.
Specific Australian dynamics include:
The Robodebt legacy — Section 12.1 covered the case; the broader political-and-social consequences have shaped Australian government-AI policy substantially. The labour-protective framing of the post-Robodebt policy framework is informative for understanding the Australian context.
Mining-and-resources labour transition — Section 9.11 covered the autonomous-mining trajectory; the labour transition has been substantial but managed over a decade-plus timeframe. The Pilbara remote-operations centres have substituted for specific roles while creating new ones; the cumulative employment has remained substantial.
Agricultural labour — Section 11.7 covered the seasonal-labour question; the PALM scheme has expanded substantially and addresses the labour-availability constraint that AI-and-automation deployment partially substitutes for.
Knowledge-work augmentation — Australian professional services have followed the broader pattern (augmentation rather than displacement); employment in legal, accounting, and consulting services has continued to grow.
Universities and research — Australian universities have substantially deployed AI (the cheating-and-plagiarism question of Section 12.2 has been particularly visible in Australia); the academic-employment dynamics have been mixed.
The 2024–2026 Australian policy response has included: the Voluntary AI Safety Standards (Section 14.8); the proposed Mandatory AI Guardrails for High-Risk AI; the Privacy Act reform with AI-specific provisions; specific labour-market protections within sectoral frameworks. The policy response is incremental but more substantive than the US response and less prescriptive than the EU response.
Malaysian labour-market dynamics. Malaysia’s labour market context differs substantially from Australia’s. The economy is dual-structured: a high-skilled urban sector (manufacturing; services; financial; healthcare) and a more-traditional rural sector (agriculture; small-and-medium enterprises). The AI deployment is concentrated in the urban sector; the rural sector is largely unaffected through 2024–2026.
Specific Malaysian dynamics include:
Manufacturing labour and the skills transition — Section 9.12 covered the Malaysian E&E cluster including ViTrox. The high-skilled manufacturing employment continues to grow; lower-skilled manufacturing has been progressively automated for decades and the AI wave is incremental in that context. The skills-transition question for the next-generation Malaysian workforce is substantial; technical-and-vocational education programs (TVET) have been expanded through 2024–2026 to address the skill-supply question.
Services and knowledge work — Malaysian financial services, healthcare, and adjacent sectors have followed the broader augmentation pattern. The KL professional-services cluster has been growing through 2024–2026 with substantial AI deployment supporting (rather than replacing) professional employment.
Migrant labour — Malaysia has substantial migrant labour (primarily from Indonesia, Bangladesh, Nepal, and the Philippines). The agricultural and manufacturing migrant-labour positions face specific automation pressures; the broader migrant-labour framework is in transition through 2024–2026 with the rights-and-conditions question being substantively contested.
Education and SPM context — Section 12.2 introduced the Malaysian SPM tutoring market that the Team Aroma worked example addresses. The broader education-AI deployment in Malaysia is at early stages; the Education Ministry has issued AI-use guidance through 2024–2026 but comprehensive policy frameworks are still developing.
Malaysian policy response has been less developed than Australian policy. The National AI Roadmap (2021) provides the strategic framework; specific labour-AI-protective provisions are at early stages. The 2024 PDPA amendment (Section 14.9) has substantial implications; sector-specific guidance from BNM (financial services), MDA (medical devices), and adjacent regulators has been developing.
The cross-regional dynamics. The unit’s KL-Melbourne dual-cohort context produces specific cross-regional labour dynamics. Australian-trained graduates often have Malaysian options (both in Malaysian operations of multinationals and in Malaysian firms); Malaysian-trained graduates have Australian options through skilled-migration pathways. The cross-regional employment market is substantial; both regions’ graduates face competition and opportunity from the other region’s labour market.
15.13 The institutional response — unions, professional associations, regulators
A specific dimension of labour-AI dynamics is the institutional response — the responses of organised labour, professional associations, and regulatory bodies. The institutional layer mediates between aggregate labour-market dynamics and individual worker outcomes.
The organised-labour response. The Hollywood strike provisions (Section 10.7) are the most-detailed organised-labour response to AI deployment. The provisions include: specific protections for writers (AI cannot serve as source material); specific protections for performers (digital replicas require consent and compensation); specific bargaining-power frameworks for ongoing AI-and-labour negotiations.
Beyond Hollywood, specific labour-organising efforts have addressed AI-related concerns:
The German IG Metall has negotiated AI-related provisions in specific sectoral agreements through 2024–2026. The UK trade-union landscape has produced specific AI-protective frameworks in collective-bargaining agreements at major employers. The US AFL-CIO has developed specific AI-related policy positions; the 2024 election cycle produced substantial labour-movement attention to AI. The Korean labour movement has produced specific AI-related collective-bargaining provisions in major manufacturing and service sectors.
The overall pattern is that organised labour, where present, can substantially shape AI deployment; the absence of organised labour in many sectors leaves the AI deployment substantially less constrained.
Professional associations. Section 11.10 covered the legal-profession response in detail; the American Bar Association’s Formal Opinion 512 (July 2024) is the most-detailed example. Other professional associations have produced parallel responses:
Medical associations — the American Medical Association, the Royal Australian College of General Practitioners, the Malaysian Medical Association, and many others have issued AI-related guidance through 2024–2026. Accounting professional bodies — the AICPA, CPA Australia, and adjacent bodies have developed AI-related professional standards. Engineering professional bodies — the IEEE, Engineers Australia, the Board of Engineers Malaysia, and others have produced AI-specific standards. Other professional bodies — lawyers, architects, designers, journalists, and many other professions have produced AI-related professional codes.
The professional-association response is less binding than legislation but substantive in shaping practice. The codes establish what professional behaviour AI deployment requires; the disciplinary frameworks enforce the codes for licensed professionals.
The regulatory response. Chapter 14 covered the regulatory response in detail. The labour-specific dimensions include the EU AI Act’s high-risk-employment provisions; the Colorado AI Act’s employment-decision provisions; NYC Local Law 144’s bias-audit requirements; the Australian Privacy Act reform’s automated-decision provisions; various adjacent frameworks. The cumulative regulatory framework is substantial; the operational implications for AI deployment in employment contexts are significant.
The collective-action problem. A specific challenge for the institutional response is the collective-action problem: individual workers’ interests in protective frameworks may differ from individual workers’ interests in immediate employment options. Protective frameworks that constrain AI deployment may slow productivity growth and economic expansion, with longer-term costs for the workers they aim to protect. The balance between protection and productivity is itself contested; institutional structures attempt to manage the balance with varying degrees of success.
The cumulative institutional response is substantial; specific frameworks address specific concerns; the broader political-and-economic dynamics produce continued evolution. The 2026–2030 trajectory will produce substantial maturation of the institutional response as the AI deployment pattern stabilises.
15.14 The 2026–2030 forward trajectory
Five trajectories define the labour-AI forward look.
Trajectory 1 — labour-market accommodation continues. The 2024–2026 pattern of aggregate employment growth despite AI deployment may continue through 2026–2030 if the augmentation-rather-than-displacement pattern persists. The pattern requires continued reinstatement effects (new occupation creation; expanded employment in AI-benefited firms) to offset specific substitution effects. The empirical evidence through 2030 will substantially shape whether the accommodation pattern holds.
Trajectory 2 — capability advancement may shift the balance. Foundation-model capability is improving; capabilities that 2026 models cannot reliably perform (complex multi-step reasoning; nuanced judgment; long-horizon planning) may become reliable in 2028–2030 models. As capability expands, the proportion of tasks that AI can substitute for may grow; the augmentation-displacement balance may shift toward more displacement. The trajectory depends substantially on whether capability advancement continues or whether technical limits constrain it.
Trajectory 3 — productivity acceleration may emerge. The J-curve productivity dynamics suggest that substantial productivity gains may emerge in 2028–2030 from the 2023–2026 deployment efforts. The acceleration would produce substantial economic benefits, with implications for wages, employment, and broader welfare. The acceleration is not certain; the 2030 macroeconomic data will substantially shape the assessment.
Trajectory 4 — institutional response maturation. The institutional response to AI will continue to develop. Organised labour, professional associations, and regulators will progressively adapt; specific protective frameworks will be refined; the cumulative institutional layer will be more substantial in 2030 than in 2026. The maturation will affect AI deployment patterns by setting the operational constraints within which deployment occurs.
Trajectory 5 — geographic and demographic concentration may produce political-economic effects. The concentration of AI labour effects in specific geographies and demographic groups produces political-economic consequences that may exceed the aggregate effects. Specific affected populations (mid-career professionals in specific roles; specific geographic clusters; specific demographic groups) may produce political mobilisation that shapes broader policy. The 2024 and 2026 election cycles in major economies have shown some labour-AI political dynamics; the 2028–2030 cycles will likely show more.
The bridge to subsequent Part III chapters: Chapter 16 develops the maturity framework that allows specific deployments to be assessed against capability and operational maturity. Chapter 17 integrates the analytical frameworks of Chapters 13–16. Chapter 18 returns to specific cases at greater synthesised depth.
The labour and economic effects of AI deployment are substantial and unfolding. The aggregate effects through 2026 have been more modest than the most-aggressive 2023 projections suggested. The distributional effects have been substantial in specific contexts. The 2026–2030 trajectory will produce more-definitive evidence on the longer-run dynamics. The framework this chapter develops — the task-based decomposition; the augmentation-vs-displacement-vs-reinstatement balance; the institutional-response layer; the policy-response landscape — provides the analytical tools for understanding both contemporary and future labour-AI dynamics.
References for this chapter
Foundational labour economics
- Acemoglu, D. and Restrepo, P. (2018). The race between man and machine: Implications of technology for growth, factor shares, and employment. American Economic Review 108(6): 1488–1542.
- Acemoglu, D. and Restrepo, P. (2019). Automation and new tasks: How technology displaces and reinstates labor. Journal of Economic Perspectives 33(2): 3–30.
- Acemoglu, D. and Restrepo, P. (2020). Robots and jobs: Evidence from US labor markets. Journal of Political Economy 128(6): 2188–2244.
- Acemoglu, D. (2024). The simple macroeconomics of AI. NBER Working Paper 32487, April 2024.
- Autor, D. H., Levy, F., and Murnane, R. J. (2003). The skill content of recent technological change: An empirical exploration. Quarterly Journal of Economics 118(4): 1279–1333.
- Autor, D. H. and Dorn, D. (2013). The growth of low-skill service jobs and the polarization of the US labor market. American Economic Review 103(5): 1553–1597.
- Goldin, C. and Katz, L. F. (2008). The Race between Education and Technology. Belknap.
Macroeconomic projections
- Goldman Sachs Global Investment Research (2023). The potentially large effects of artificial intelligence on economic growth. March 2023.
- Frey, C. B. and Osborne, M. A. (2017). The future of employment: How susceptible are jobs to computerisation? Technological Forecasting and Social Change 114: 254–280.
- McKinsey Global Institute (2024). Generative AI in the Workplace.
- International Monetary Fund (2024). Gen-AI: Artificial Intelligence and the Future of Work.
- Boston Consulting Group (2024). AI labor-market analysis.
Productivity and field studies
- Brynjolfsson, E., Li, D., and Raymond, L. R. (2023). Generative AI at work. NBER Working Paper 31161; subsequently published in Quarterly Journal of Economics (2025).
- Peng, S., Kalliamvakou, E., Cihon, P., and Demirer, M. (2023). The impact of AI on developer productivity: Evidence from GitHub Copilot. arXiv:2302.06590.
- Noy, S. and Zhang, W. (2023). Experimental evidence on the productivity effects of generative artificial intelligence. Science 381(6654): 187–192.
- Eloundou, T., Manning, S., Mishkin, P., and Rock, D. (2023). GPTs are GPTs: An early look at the labor market impact potential of large language models. arXiv:2303.10130; updated 2024.
- Brynjolfsson, E. and Hitt, L. M. (2003). Computing productivity: Firm-level evidence. Review of Economics and Statistics 85(4): 793–808.
- Solow, R. M. (1987). We’d better watch out. New York Times Book Review, 12 July 1987.
Reskilling and labour-market policy
- Card, D., Kluve, J., and Weber, A. (2018). What works? A meta analysis of recent active labor market program evaluations. Journal of the European Economic Association 16(3): 894–931.
- Workforce Innovation and Opportunity Act (US, 2014, with subsequent modifications).
- European Social Fund Plus framework (2021–2027).
- Singapore SkillsFuture program documentation.
Universal Basic Income
- OpenResearch (2024). UBI Pilot Study Report.
- Hoynes, H. and Rothstein, J. (2019). Universal basic income in the United States and advanced countries. Annual Review of Economics 11: 929–958.
Australian and Malaysian context
- Australian Bureau of Statistics (2024, 2025, 2026). Labour force statistics and reports.
- Department of Employment and Workplace Relations (Australia, 2024). Future Skills Forecast.
- Department of Statistics Malaysia (2024, 2025, 2026). Labour force statistics.
- Ministry of Human Resources Malaysia (2024). National TVET Policy.
Sector-specific labour analyses
- Writers Guild of America (2023). 2023 MBA Memorandum of Agreement (with AI provisions).
- Screen Actors Guild–American Federation of Television and Radio Artists (2023). 2023 TV/Theatrical Memorandum of Agreement.
- Stanford CodeX (2024). AI in Law programme reports.
- American Bar Association (2024). Formal Opinion 512 on AI use by lawyers.