Economic × AI · Linh Truong
◆ Personal notes · May 2026

Economic Transformation in the AI Era

I put this together to make sense of what AI is doing to production, labor, capital, and growth — and to turn that into decisions I can actually use. It is not a forecast. It is the economic map I am working from.

The question behind this note: where does the surplus from cheap intelligence actually go — and how do I position myself, my work, and my bets before the diffusion curve steepens? What follows is my read of the forces, the frictions, and the levers worth pulling.
📍 Scope: Macro & firm-level 📅 Horizon: 2022 → 2035 🧭 Stance: Diffusion beats capability ✍️ By: Linh Truong
~15–25%
Mid-range estimate for cumulative global labor-productivity lift from generative + agentic AI through the 2030s
60–80%
Share of occupations where a meaningful fraction of tasks face AI augmentation or automation
$2–4T
Annual value-creation potential across sales, software, service, R&D, and ops at scale
3 layers
Where I see value concentrating: compute & models · infrastructure & data · applications & workflow
01 · What I'm tracking

The economic story in one page

AI is a general-purpose technology (GPT) — like steam, electricity, and the internet. The impact does not come from any single model; it comes from complementary reinvention: new workflows, organizational forms, skills, and institutions built around the capability. The transition is uneven, contested, and policy-sensitive — which is why I watch diffusion as closely as capability.

Mechanism

From tasks to systems

Value first appears as task-level automation/augmentation, then compounds as firms redesign processes end-to-end. Agentic AI shifts the unit of work from "answer a prompt" to "complete a workflow."

Timing

The J-curve / productivity paradox

GPTs historically show a lag then surge: measured productivity dips while firms invest in intangibles (data, skills, process), then accelerates. Expect uneven gains 2024–2028, broader diffusion after.

Distribution

Who captures the gains

Outcomes hinge on policy & market structure: gains can be broadly shared (augmentation, competition, reskilling) or concentrated (winner-take-most platforms, capital-biased automation).

🎯

What I keep coming back to: The economic question is no longer "can AI do the task?" but "how fast can organizations and societies reorganize around it — and on what terms is the surplus shared?" I put most of my strategic weight on the diffusion and adaptation bottleneck, not just model capability.

02 · The Big Picture

AI economic transformation — system map

How capability flows through the economy — from inputs, into firms and the labor/capital mix, to output and growth, and finally to how the surplus is distributed. The feedback loops are what I watch most closely: they can accelerate the whole system or stall it.

FIG 1

Inputs → Firms → Output → Distribution (with feedback loops)

1 · INPUTS 2 · FIRMS & WORK 3 · OUTPUT & GROWTH 4 · DISTRIBUTION Compute & Energy GPUs · data centers · power Data proprietary · web · sensor Models & Algorithms frontier · open · agents Talent & Capital researchers · investment Task Automation replace routine work Task Augmentation amplify human output Process Redesign new workflows & orgs New Products/Markets AI-native business models Productivity ↑ output per hour Lower Costs / Prices cheaper cognition Innovation Rate ↑ faster R&D & discovery Wages & Jobs labor share & skills Profits & Capital returns to owners Consumer Surplus access & quality Public Revenue tax base & services + Reinvestment loop: profits → compute, data, talent − Adoption friction: trust, skills, regulation, integration ⚙ Diffusion bottleneck intangible investment lag
How I read it left→right: capability is produced from inputs, absorbed by firms via automation/augmentation, converted to output (productivity, lower prices, innovation), then distributed. The teal loop shows how surplus is reinvested to accelerate; the red loop shows the frictions that slow diffusion. Most of the leverage I see — policy and corporate — sits at the diffusion bottleneck.
03 · What's Driving It

The forces behind the shift

Six reinforcing drivers explain why this wave is different from prior automation cycles — and why the curve is steepening now.

Capability

Frontier & agentic models

Models moved from narrow tools to general reasoning and multi-step agents that plan, use tools, and complete tasks — expanding the set of automatable cognitive work.

Economics

Collapsing cost of cognition

Inference cost per useful task has fallen by orders of magnitude. When a capability gets ~10× cheaper, demand and new use cases expand non-linearly.

Access

Natural-language interface

No-code, conversational access means non-engineers can deploy automation, collapsing the gap between idea and implementation.

Infrastructure

Compute & energy build-out

Hyperscale capex, specialized chips, and data-center/energy expansion create the physical substrate — and a new strategic resource constraint.

Capital

Investment flywheel

Record private + corporate investment funds talent, compute, and applications, compressing iteration cycles across the stack.

Diffusion

Embedded in tools of work

AI ships inside the software people already use (docs, code, CRM, search), lowering the adoption barrier from "buy a system" to "turn on a feature."

Counter-forces I watch: reliability & hallucination, data governance & IP, security, regulatory uncertainty, integration with legacy systems, organizational change capacity, energy/compute limits, and public trust. Capability is rarely the binding constraint — absorption is.

04 · Adoption Curve

Four phases of diffusion (2022 → 2035)

A stylized S-curve: experimentation, then production deployment, then process & market restructuring, then institutional reset. The economy moves through these unevenly across firms and regions.

FIG 2

Capability vs. economic impact — the J-curve / S-curve

Economic impact → Time → 2022–24 2024–27 2027–30 2030–35 Adoption (S-curve) measured productivity (J-curve) intangible investment dip ① Experiment ② Deploy ③ Restructure ④ Institutionalize
Adoption races ahead of measured productivity. Early on, firms pour resources into data, skills, and process redesign (intangibles) before gains show up in the statistics — the classic "productivity J-curve." I try to plan for where the curve is going, not where the headlines are.
PHASE 1 · 2022–24

Experimentation

Pilots, copilots, and proofs-of-concept. Productivity gains localized to individual tasks; ROI hard to measure; "shadow AI" spreads bottom-up.

PHASE 2 · 2024–27

Production deployment

Agents move into core workflows. Function-level redesign (support, code, marketing, ops). Early winners separate from laggards; data & governance become differentiators.

PHASE 3 · 2027–30

Restructuring

Org charts, supply chains, and business models reshape around AI. New AI-native entrants pressure incumbents. Productivity surge becomes visible in aggregate stats.

PHASE 4 · 2030–35

Institutionalization

Labor markets, education, tax, and safety nets adapt. New norms, standards, and regulation settle. The "new normal" of an AI-augmented economy.

05 · Where It Lands

Sector impact heat map

Exposure varies by how much of a sector's value is cognitive/digital vs. physical/regulated. High exposure ≠ high readiness — the gap between the two is where the disruption (and opportunity) concentrates.

TBL 1

Exposure, near-term ROI, and dominant mode by sector

SectorAI ExposureNear-term ROIDominant modeWhat changes first
Software & ITVery HighHighAugment → AutomateCode generation, testing, agentic dev, ops
Financial servicesVery HighHighAugmentResearch, underwriting, compliance, advisory
Professional & legalHighMediumAugmentDrafting, review, due diligence, knowledge work
Media, marketing & designVery HighHighAutomate + CreateContent, creative, personalization at scale
Customer service & BPOVery HighHighAutomateTier-1 resolution, agent assist, deflection
HealthcareHighMediumAugmentDocumentation, imaging, triage, discovery
EducationHighMediumAugment + CreateTutoring, content, assessment, admin
ManufacturingMediumMediumAugmentDesign, quality, predictive maintenance, robotics
Retail & e-commerceHighHighAugment + AutomateSearch, recommendations, demand & supply ops
Construction & agricultureLow–MedLowAugmentPlanning, monitoring, robotics (slower physical)
Government & publicMediumMediumAugmentService delivery, casework, fraud, analytics
Pattern: cognitive + digital + low-regulatory-friction sectors transform first; physical + safety-critical + highly-regulated sectors lag but eventually see deep change via robotics and verified systems. Legend: Low Medium High Very High.
06 · The Labor Question

Automate · Augment · Create

AI doesn't simply "destroy jobs." It restructures tasks within jobs. Net employment depends on the balance of three forces — displacement, productivity-driven reinstatement, and the creation of entirely new work.

Displacement

Tasks automated

Routine cognitive and predictable tasks are substituted. Roles heavy in these tasks shrink or consolidate. Most exposed: data entry, basic content, tier-1 support, routine analysis, simple coding.

Augmentation

Tasks amplified

Humans + AI outperform either alone. Output per worker rises; the skill premium shifts toward judgment, orchestration, and verification. Often raises demand for the augmented role.

Creation

New work appears

New roles, tasks, and entire categories emerge — AI orchestration, eval & safety, data & context engineering, plus expansion of human-premium work (care, trust, in-person).

FIG 3

The skills barbell — where human value migrates

hollowing-out zone HUMAN-PREMIUM (rising) Care · trust · in-person Manual dexterity Relationships & negotiation AI-LEVERAGED (rising) Judgment & strategy Orchestration & design Verification & ethics routine cognitive · predictable · repetitive (falling) value migrates to the ends of the barbell
Demand polarizes toward two ends: uniquely-human work (physical, relational, care) and AI-leveraged high-judgment work. The squeezed middle — routine cognitive work — is where reskilling and transition policy matter most.
Net effect

Why the headline number is uncertain

  • Displacement effect: tasks done by capital, not labor → fewer hours for some roles.
  • Productivity effect: cheaper output → lower prices → higher demand → more total work.
  • Reinstatement effect: new tasks where labor has comparative advantage are created.
  • Composition effect: winners and losers differ by skill, sector, geography, and age.
The transition risk

Speed & distribution, not totals

  • Even if long-run employment holds, the transition can be painful and uneven.
  • Adjustment costs fall hardest on mid-skill workers and exposed regions.
  • Wage effects may precede job effects (compression before displacement).
  • Policy levers (reskilling, portable benefits, wage insurance) shape the human cost.
07 · The Economics

How AI shows up in growth

Through the lens of growth accounting, AI raises output via three channels — and there's a debate between "big productivity boom" optimists and "limited / lagged" skeptics.

Channel 1

Capital deepening

Investment in AI compute, software, and equipment raises capital per worker. Big near-term GDP contribution from the build-out itself (data centers, chips, energy).

Channel 2

Labor productivity

The same workers produce more via augmentation and automation of tasks — the central long-run channel as AI diffuses into everyday workflows.

Channel 3

TFP & the "ideas" effect

If AI accelerates R&D and discovery itself ("automating the production of ideas"), it could lift total factor productivity and even the growth rate, not just the level.

FIG 4

The optimist–skeptic spectrum on aggregate impact

Skeptic Measured Optimist Transformative "Limited & lagged;few exposed tasks" "Real but gradual;~0.5–1.5pp/yr" "Broad boom asagents scale" "Growth-rate shift;AI does R&D" Most credible base case: real, compounding gains that arrive unevenly — closer to "Measured/Optimist" — with wide error bars.
The honest answer is a range. The economically interesting variable is not peak capability but how quickly cheaper cognition diffuses into the 80% of the economy that is services, and whether AI begins to compound by accelerating its own R&D inputs.
📐

My mental model: Output = f(Capital, Labor, Ideas/TFP). AI pushes on all three. The level effect (one-time productivity boost) looks near-certain; the high-stakes possibility is a rate effect — a persistently faster growth trajectory if AI durably accelerates innovation. I plan for the level effect and stay alert to the rate effect.

08 · The Strategy

Playbooks for three actors

The transition is shaped by choices at three levels — policy, enterprise, and individual. I use this stack to check whether the levers are aligned or working against each other.

FIG 5

The strategy stack — interlocking levers

POLICY rules of the game ENTERPRISE competitive advantage & adoption INDIVIDUAL skills, agency & livelihood sets guardrails & safety nets ↑ creates & captures value → supplies & absorbs change →
Each layer constrains and enables the others. Misalignment (e.g., fast automation without reskilling, or heavy regulation without adoption support) produces backlash and forgone gains; alignment produces broadly-shared growth.
For Policymakers

Steer diffusion & share gains

  • Invest in adoption: diffusion programs for SMEs & public sector, not just frontier research.
  • Reskilling at scale: fund lifelong learning, fast credentials, apprenticeships.
  • Modernize safety nets: portable benefits, wage insurance, transition support.
  • Competition policy: keep compute, data & model markets contestable; interoperability.
  • Smart regulation: risk-based, pro-innovation guardrails (safety, transparency, liability).
  • Public infrastructure: compute access, open data, energy, digital public goods.
  • Fiscal foresight: protect the tax base; plan for distributional shifts.
For Enterprises

Redesign, don't just bolt on

  • Start with workflows, not tools: target high-volume, high-value processes end-to-end.
  • Data & context as moat: proprietary data, retrieval, and feedback loops compound.
  • Augment first, automate deliberately: redeploy freed capacity to growth.
  • Build the operating model: governance, evals, human-in-the-loop, change management.
  • Talent strategy: upskill broadly; hire orchestration & AI-product skills.
  • Measure ROI honestly: baseline, A/B, track quality & risk, not just adoption.
  • Manage risk: security, IP, model/vendor concentration, reliability, compliance.
For Individuals

Become AI-leveraged

  • Move up the barbell: deepen judgment, creativity, relationships, and domain expertise.
  • Learn to orchestrate AI: prompting, tool use, verification, and workflow design.
  • Pair AI with a moat: rare human skills + AI leverage beat either alone.
  • Stay liquid: portfolio of skills; bias toward roles that grow with AI.
  • Verify, don't outsource thinking: keep the human judgment loop sharp.
  • Continuous learning: treat skill renewal as a permanent habit, not an event.
09 · What Could Go Wrong

Risk & governance map

A transformation this broad creates systemic risks. I am not arguing for halting progress — but for internalizing externalities and keeping the transition legitimate and stable.

FIG 6

Risk quadrant — likelihood × severity

Severity → Likelihood (near-term) → monitor manage actively watch tail risks act now Jobtransition Inequality Marketconcentration Misinfo /trust Security /cyber Energy /compute Systemic /safety tail Bias /fairness
Bottom-right ("act now") = high likelihood + high severity: job transition and misinformation/trust. Top-left ("watch") = lower likelihood but catastrophic if realized: systemic safety tail risks. Governance should be proportionate to each cell.
Distribution

Inequality & concentration

Gains may accrue to capital owners, top talent, and dominant platforms — widening wealth/wage gaps and concentrating market & geopolitical power.

Society

Trust, truth & cohesion

Synthetic media, manipulation, and erosion of shared facts threaten institutions, democracy, and consumer trust.

Systemic

Safety, control & resilience

Reliability failures, automation fragility, model-monoculture risk, and longer-horizon advanced-AI safety concerns.

Resource

Energy & compute

Compute scarcity, data-center energy/water demand, and supply-chain concentration become strategic and environmental constraints.

Rights

Privacy, IP & data

Training data provenance, copyright, surveillance, and personal-data governance remain contested and litigated.

Governance

The response toolkit

Risk-based regulation, evals & standards, transparency, liability rules, antitrust, reskilling funds, and international coordination.

10 · Make It Actionable

Strategic roadmap — three horizons

The sequence I use when turning macro insight into action. Build foundations before scaling; scale before restructuring.

Horizon 1 · Foundations0–12 months
  • Map exposure: inventory tasks/workflows by value & AI-fit.
  • Set governance: data, security, acceptable-use, human-in-loop.
  • Run targeted pilots: 2–3 high-ROI workflows with clear baselines.
  • Upskill broadly: AI literacy for everyone; champions per team.
  • Fix data foundations: access, quality, retrieval, feedback capture.
Horizon 2 · Scale1–3 years
  • Productionize winners: move pilots to core workflows with SLAs.
  • Deploy agents: multi-step automation with oversight & evals.
  • Redeploy capacity: shift freed time to growth, quality, new offers.
  • Build the moat: proprietary data + feedback loops compounding.
  • Measure & iterate: ROI, quality, risk dashboards; kill what fails.
Horizon 3 · Reinvent3–5+ years
  • Redesign the org: structure, roles & incentives around AI.
  • New business models: AI-native products, pricing, and markets.
  • Ecosystem position: partnerships, platforms, standards.
  • Workforce of the future: human-premium + AI-leveraged roles.
  • Resilience: diversify models/vendors, energy, and risk controls.
🧭

Sequencing principle I follow: Augment → Automate → Reinvent. Most failed AI programs I have seen skip straight to automation or to "transformation" decks without the data foundations, governance, and trust that make either stick. Earn the right to restructure by first proving value and building capability.

11 · Where This Goes

2030–2035 scenarios

Two axes dominate the outcome space: how fast capability diffuses (pace) and how broadly gains are shared (distribution). Four plausible worlds result.

FIG 7

Scenario matrix — pace × distribution

FAST diffusion SLOW diffusion CONCENTRATED BROADLY SHARED ⚡ Winner-Take-Most Fast gains captured by few platforms & capital owners. High growth, high inequality, political backlash risk. Watch: concentration, labor share ↓ 🌍 Broad Prosperity Fast diffusion + competition + reskilling. Productivity boom widely shared; new work created; rising real incomes. The goal: align policy + enterprise + skills 🧱 Stalled & Captured Hype fades into slow adoption; gains still accrue narrowly. Forgone growth + entrenched incumbents. Worst of both. Watch: regulatory capture, low diffusion 🐢 Gradual & Inclusive Slow but steady, broadly-shared gains. Lower disruption, lower upside; time to adapt institutions. "Muddle through." Likely default for cautious economies
My aim is to push the system toward Broad Prosperity (top-right): accelerate diffusion and widen participation. The levers that move us there are the policy, enterprise, and individual playbooks in §08 working together.
12 · Track Progress

Metrics that matter

Signals I watch to know which scenario we are drifting toward — at macro and organizational levels.

Macroeconomics

National & global signals

Labor productivity growthOutput/hour trend; is the J-curve turning up?
AI adoption rate% of firms/workers using AI in core work
Labor share of incomeIs the surplus reaching workers?
Wage dispersionPremiums by skill; middle hollowing?
Job churn & new rolesCreation vs. destruction; reemployment speed
Market concentrationCompute, models, downstream markets
Compute & energyCapacity, price, emissions intensity
Micro / Organization

Enterprise dashboard

Workflow ROICost/quality/speed vs. baseline, by use case
Adoption depthActive use in core workflows (not just logins)
Quality & error rateAccuracy, escalations, rework, incidents
Cycle timeEnd-to-end process time reduction
Capacity redeployedFreed hours → growth vs. pure cost cut
Skill coverage% workforce AI-fluent; orchestration roles
Risk postureSecurity, compliance, model/vendor concentration
⚠️

Vanity-metric trap: "seats deployed" or "prompts per day" measure activity, not value. I anchor on outcome metrics (quality-adjusted output, cycle time, ROI, redeployed capacity) and always compare against a real baseline.

13 · Remember This

Ten things I keep on my desk

Not commandments — reminders I re-read when the macro noise gets loud.

1 · AI is a general-purpose technology

Impact comes from complementary reinvention — workflows, skills, institutions — not the model alone.

2 · Diffusion is the bottleneck

Absorption capacity, not capability, governs the pace. That's where I think strategy has the most leverage.

3 · Expect a J-curve

Investment lag before the productivity surge. I don't mistake the dip for failure.

4 · Tasks, not jobs

AI restructures task bundles. Net employment depends on displacement vs. reinstatement vs. creation.

5 · Value polarizes (barbell)

Human-premium and AI-leveraged work rise; routine cognitive work gets squeezed.

6 · Distribution is a choice

Policy and market structure decide whether gains are broadly shared or concentrated.

7 · Sequence: Augment → Automate → Reinvent

Earn the right to restructure by first proving value and building capability.

8 · Data & context are the moat

Proprietary data plus feedback loops compound advantage more than model access alone.

9 · Govern proportionately

Match guardrails to likelihood × severity; internalize externalities without stalling diffusion.

10 · Aim for Broad Prosperity

Accelerate diffusion and widen participation — the only quadrant that's both high-growth and stable.

14 · References & Sources

Where the ideas in this note come from

Annotated bibliography behind the GPT thesis, system map, adoption J-curve, sector heat map, task-based labor framework, growth channels, strategy stack, risk quadrant, scenario matrix, and KPI dashboard. Section tags (e.g. §06) show where each source is used. Diagrams and operating rhythms are my synthesis unless noted.

Scope. Synthesis of labor economics, growth accounting, industry reports, and field experiments (May 2026). Hero-strip figures (e.g. ~15–25% productivity lift, $2–4T value, 60–80% occupational exposure) are mid-range ranges drawn from the sources below—not forecasts for any firm or economy. Not legal, financial, employment, or investment advice.

Citations are numbered continuously [1]–[n] within this section.

General-purpose technology, diffusion & the productivity J-curve (§01, §03–§04, FIG 2)

  1. David, P. A., “The Dynamo and the Computer: An Historical Perspective on the Modern Productivity Paradox.” American Economic Review (Papers & Proceedings), 1990. Historical parallel between electricity and IT: capability diffuses before it shows up in the statistics—background for the diffusion bottleneck in FIG 1. doi.org/10.1257/aer.90.2.355 — §01, §04.
  2. Bresnahan, T. F. & Trajtenberg, M., “General Purpose Technologies: ‘Engines of Growth’?” Journal of Econometrics, 1995. Formal GPT framework; complementary investments and long adoption lags underpin §01 “complementary reinvention.” doi.org/10.1016/0304-4076(94)01598-T — §01.
  3. Brynjolfsson, E., Rock, D., & Syverson, C., “The Productivity J-Curve: How Intangibles Complement General Purpose Technologies.” American Economic Journal: Macroeconomics, 2021. Intangible investment lag explains measured productivity dipping before surging—core to FIG 2 and §04 phases. doi.org/10.1257/mac.20180386 — §01, §04, §12.
  4. Brynjolfsson, E., Rock, D., & Syverson, C., “Artificial Intelligence and the Modern Productivity Paradox: A Clash of Expectations and Statistics.” NBER Working Paper 24001, 2018. Applies the J-curve logic to AI specifically; supports “capability races ahead of measured impact.” nber.org/papers/w24001 — §01, §04.
  5. Brynjolfsson, E. & McAfee, A., The Second Machine Age: Work, Progress, and Prosperity in a Time of Brilliant Technologies. W. W. Norton, 2014. Digital GPT wave, complementary skills, and “race with the machine” framing in §01 and §08. — §01, §08.

Task-based automation & labor-market mechanics (§01, §06, FIG 3)

  1. Acemoglu, D. & Restrepo, P., “The Race between Man and Machine: Implications of Technology for Growth, Factor Shares, and Employment.” American Economic Review, 2018. Displacement, productivity, and reinstatement effects—the three-force balance in §06 cards. doi.org/10.1257/aer.20160696 — §06.
  2. Acemoglu, D. & Restrepo, P., “Automation and New Tasks: How Technology Displaces and Reinstates Labor.” Journal of Economic Perspectives, 2019. New-task creation as offset to automation; supports “tasks, not jobs” framing. doi.org/10.1257/jep.33.2.3 — §06, §13.
  3. Autor, D. H., “Why Are There Still So Many Jobs? The History and Future of Workplace Automation.” Journal of Economic Perspectives, 2015. Historical pattern: automation shifts work rather than eliminating it net—§06 “creation” column. doi.org/10.1257/jep.29.3.3 — §06.
  4. Autor, D. H., Levy, F., & Murnane, R. J., “The Skill Content of Recent Technological Change: An Empirical Exploration.” Quarterly Journal of Economics, 2003. Routine vs. non-routine task framework underlying the skills barbell in FIG 3. doi.org/10.1162/00335500360535172 — §06, FIG 3.
  5. Autor, D. H., “The Work of the Past, Work of the Future.” AEA Papers & Proceedings, 2019. Task polarization and rising value of social–analytic work at the AI-leveraged end of the barbell. doi.org/10.1257/pandp.20191110 — §06, FIG 3.

Occupational exposure, sectors & adoption (§05, hero stats, TBL 1)

  1. Eloundou, T., Manning, S., Mishkin, P., & Rock, D., “GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models.” arXiv, 2023. ~80% of U.S. workers in occupations with ≥10% of tasks LLM-exposed; GPT-as-GPT argument in §01. arxiv.org/abs/2303.10130 — hero, §01, §05.
  2. Felten, E., Raj, M., & Seamans, R., “Occupational, Industry, and Geographic Exposure to Artificial Intelligence: A Novel Dataset and Its Potential Uses.” Strategic Management Journal, 2021. AIOE dataset linking AI applications to O*NET abilities—methodological cousin to exposure mapping in TBL 1. doi.org/10.1002/smj.3286 — §05.
  3. McKinsey Global Institute, The Economic Potential of Generative AI: The Next Productivity Frontier. June 2023. $2.6–4.4T annual value across use cases; 60–70% of employee time in automatable activities—feeds hero $2–4T strip and sector ROI column. mckinsey.com — hero, §05, §07.
  4. McKinsey Global Institute, The Future of Work in America: People and Places, Today and Tomorrow. 2019. Occupational shift patterns in knowledge-intensive sectors—context for sector heat-map ordering. mckinsey.com/mgi — §05.
  5. Stanford HAI, 2025 AI Index Report — Economy chapter. 2025. Investment, adoption, and corporate AI deployment trends; supports §03 drivers and §12 KPIs. hai.stanford.edu/ai-index — §03, §12.

Field evidence on augmentation & workflow productivity (§06, §08, §10)

  1. Brynjolfsson, E., Li, D., & Raymond, L. R., “Generative AI at Work.” Quarterly Journal of Economics, 2025 (NBER WP 31161, 2023). Customer-support agents: +14% productivity on average, largest gains for less-experienced workers—§06 augmentation card. doi.org/10.1093/qje/qjae044 — §06, §08.
  2. Dell’Acqua, F. et al., “Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality.” Harvard Business School Working Paper 24-013, 2023. BCG consultant experiment; “jagged frontier” where AI helps inside competence and hurts outside—§08 “augment first” and §09 trust risks. hbs.edu — §08, §09.
  3. Noy, S. & Zhang, W., “Experimental Evidence on the Productivity Effects of Generative Artificial Intelligence.” Science, 2023. Writing tasks: ~40% faster, ~18% higher quality; compresses skill gaps—§06 augmentation and §08 individual playbook. doi.org/10.1126/science.adh2586 — §06, §08.
  4. Peng, S. et al., “The Impact of AI on Developer Productivity: Evidence from GitHub Copilot.” Microsoft Research, 2023. Controlled experiment: 55.8% faster task completion with AI coding assistant—§05 software row and §08 pilots. microsoft.com/research — §05, §08, §10.
  5. NIST, Artificial Intelligence Risk Management Framework (AI RMF 1.0). U.S. Department of Commerce, 2023. Govern–map–measure–manage cycle for enterprise AI operating models in §08–§10. nist.gov/ai-rmf — §08, §09, §10.

Macro impact estimates & growth accounting (§07, hero stats, FIG 4)

  1. Goldman Sachs Global Investment Research, “Generative AI Could Raise Global GDP by 7%.” 2023. ~7% cumulative GDP lift and ~1.5pp productivity growth over a decade—optimist end of FIG 4 spectrum. goldmansachs.com — hero, §07, FIG 4.
  2. PwC, Sizing the Prize: PwC’s Global Artificial Intelligence Study. 2017. Up to $15.7T global GDP contribution by 2030 from AI—long-horizon macro anchor cited alongside newer gen-AI estimates. pwc.com — hero, §07.
  3. IMF, Gen-AI: Artificial Intelligence and the Future of Work (Staff Discussion Note SDN/2024/001). January 2024. Cross-country AI exposure, inequality channels, and policy toolkit—§07 distribution and §08 policy column. imf.org — §07, §08.
  4. IMF, Broadening the Gains from Generative AI: The Role of Fiscal Policies (Staff Discussion Note, June 2024). Fiscal levers for shared gains—§08 reskilling and safety-net bullets. imf.org — §08, §11.
  5. OECD, Employment Outlook 2023 — AI, productivity and job quality chapter. 2023. OECD-wide labor-market and productivity framing for §07 channels and §12 macro KPIs. oecd.org/employment-outlook — §07, §12.

Endogenous growth, the “ideas” channel & the skeptic view (§07, FIG 4)

  1. Aghion, P., Jones, B. F., & Jones, C. I., “Artificial Intelligence and Economic Growth.” NBER Working Paper 23928, 2017; chapter in The Economics of Artificial Intelligence: An Agenda, 2019. AI automating idea production and possible growth-rate effects—Channel 3 in §07. nber.org/papers/w23928 — §07.
  2. Romer, P. M., “Endogenous Technological Change.” Journal of Political Economy, 1990. Ideas as non-rival inputs—intellectual foundation for TFP / “ideas” channel. doi.org/10.1086/261725 — §07.
  3. Jones, C. I., “R&D-Based Models of Economic Growth.” Journal of Political Economy, 1995. Research labor and knowledge accumulation—growth-accounting backdrop for §07 mental model. doi.org/10.1086/262016 — §07.
  4. Acemoglu, D., “The Simple Macroeconomics of AI.” NBER Working Paper 32487, 2024; Economic Policy, 2025. Task-based upper bound on TFP gains (~0.5–0.7% over 10 years)—“Measured/Skeptic” pole of FIG 4. nber.org/papers/w32487 — §07, FIG 4.
  5. Gordon, R. J., The Rise and Fall of American Growth: The U.S. Standard of Living since the Civil War. Princeton University Press, 2016. Secular stagnation / limited frontier argument—context for the left side of FIG 4. — §07, FIG 4.

Labor, skills, augmentation vs. automation globally (§06, §08, §12)

  1. International Labour Organization, Generative AI and Jobs: A Global Analysis of Potential Effects on Job Quantity and Quality (Working Paper 96). 2024. Global exposure; augmentation more likely than full automation—balances §06 three-force cards. ilo.org — §06.
  2. World Economic Forum, The Future of Jobs Report 2025. 2025. Emerging roles (AI orchestrators, evaluators, ethics leads) and skill churn—§06 creation column and §08 talent strategy. weforum.org — §06, §08, §10.
  3. Acemoglu, D. & Johnson, S., “Rebalancing AI.” Finance & Development, IMF, December 2023. Automation-heavy vs. labor-complementing AI paths—§06 transition risk and §11 distribution axis. imf.org/fandd — §06, §11.
  4. Acemoglu, D. & Restrepo, P., “Tasks, Automation, and the Rise in US Wage Inequality.” Econometrica, 2022. Task displacement and wage structure—§06 barbell squeeze and §09 inequality quadrant. doi.org/10.3982/ECTA19416 — §06, §09.

Distribution, labor share & market concentration (§07, §09, §11–§12)

  1. Karabarbounis, L. & Neiman, B., “The Global Decline of the Labor Share.” Quarterly Journal of Economics, 2014. Falling labor share since the 1980s—§12 “labor share of income” KPI and §11 concentration scenarios. doi.org/10.1093/qje/qju032 — §11, §12.
  2. Acemoglu, D. & Restrepo, P., “The Wrong Kind of AI? Artificial Intelligence and the Future of Labour Demand.” Cambridge Journal of Regions, Economy and Society, 2020. “So-so automation” that displaces without large productivity gains—§09 inequality card and FIG 6. doi.org/10.1093/cjres/rsz022 — §09.
  3. Autor, D. H. et al., “The Fall of the Labor Share and the Rise of Superstar Firms.” Quarterly Journal of Economics, 2020. Market concentration and capital-biased returns—§09 market-concentration bubble and §11 Winner-Take-Most quadrant. doi.org/10.1093/qje/qjaa006 — §09, §11.

Compute, energy & infrastructure constraints (§03, §09, §12)

  1. International Energy Agency, Energy and AI (special report). 2025. Data-center electricity ~415 TWh (2024), AI as primary growth driver, regional concentration—§03 infrastructure driver and §09 energy/compute risks. iea.org/reports/energy-and-ai — §03, §09, §12.
  2. International Energy Agency, Electricity 2024: Analysis and Forecast to 2026. 2024. Data-center load could exceed 1,000 TWh by 2026—§09 resource-risk card. iea.org/reports/electricity-2024 — §09, §12.
  3. Stanford HAI, 2025 AI Index Report — Technical Performance & Economy sections. 2025. Inference cost trends, model concentration, and capex flywheel—§03 “collapsing cost of cognition” and compute KPIs. hai.stanford.edu (PDF) — §03, §12.

Governance, regulation & risk management (§08–§10, FIG 6)

  1. OECD, OECD AI Principles. 2019 (updated 2024). Human-centered values, transparency, accountability—§08 smart-regulation bullet and §09 governance toolkit. oecd.org/ai-principles — §08, §09.
  2. European Parliament & Council, Regulation (EU) 2024/1689 (Artificial Intelligence Act). 2024. Risk-based tiers, high-risk obligations, GPAI rules—§08 enterprise compliance and §09 rights/IP card. eur-lex.europa.eu — §08, §09.
  3. Goldman Sachs Global Investment Research, “AI May Start to Boost US GDP in 2027.” 2024. Adoption lag vs. investment surge—timeline context for §04 phases and §10 Horizon 2. goldmansachs.com — §04, §10.

Author synthesis & primary note

  1. Truong, L., Economic Transformation in the AI Era — personal working notes. May 2026. Original diagrams (FIG 1–7, TBL 1), system map, scenario matrix, strategy stack, risk quadrant, three-horizon roadmap, and ten desk principles. LinhTruong.com — all sections.
  2. Truong, L., companion notes: Global Economic Transformation in the AI Era and Geopolitics Transformation in the AI Era. Cross-border diffusion, compute geopolitics, and global scenario framing referenced in §02 loops and §11 Broad Prosperity aim. Same author collection. — §02, §11.
📎

Before you quote externally: The hero-strip ranges (~15–25% productivity, 60–80% occupational exposure, $2–4T value) blend McKinsey (2023), Goldman Sachs (2023–24), Eloundou et al. (2023), and Acemoglu (2024)—they are not universal forecasts. The adoption phases in §04 compress firm-level heterogeneity into one S-curve. Re-read primary sources and your org’s policy before citing figures or deploying AI in regulated workflows.

↑ Top