Autonomous ‘agentic’ AI can cut time and costs in administrative and monitoring tasks across ageing health systems, but its value depends far more on payment rules, regulation and interoperable data than on algorithmic prowess; claims that agentic systems will materially ease ageing-driven fiscal strain remain unproven.

Main Finding

Agentic AI — defined as autonomous, goal-directed systems capable of multi-step workflow coordination — can deliver measurable short-term productivity gains in ageing health systems (notably administrative automation, scheduling, decision support, and remote monitoring). However, implementation outcomes depend far more on institutional conditions (governance, reimbursement, data/interoperability, workforce incentives, legal frameworks) than on raw algorithmic performance. Evidence strength is inversely correlated with intervention complexity: simpler, narrowly scoped interventions show the strongest empirical support; complex, cross-sector agentic deployments (e.g., autonomous care agents coordinating long-term care across providers and payers) remain weakly supported. Macro-level claims that agentic AI will substantially moderate ageing-driven fiscal pressures are empirically unvalidated.

Key Points

• Demographic and fiscal context

  • Advanced economies face steep ageing pressures: multiple countries will see 65+ shares of 30–40% by 2050.
  • Ageing-related public expenditures absorb up to ~18% of GDP in the most affected economies.
  • Demographic ageing is projected to cut annual GDP growth by ~0.3–1.2 percentage points by 2035.
  • Long-term care (LTC) challenges: severe workforce shortfalls and per-capita elderly care costs 3–5× those of working-age cohorts.

• Evidence synthesis (81 sources, 2020–2025)

  • Review types: RCTs, quasi-experimental studies, implementation pilots, case studies, policy analyses, and macro/micro modeling.
  • Stronger evidence for low-complexity interventions: administrative automation, scheduling, targeted decision support, remote monitoring — consistent productivity/time-savings and process improvements.
  • Moderate evidence for workforce augmentation tools (clinician decision augmentation, assisted telecare) — benefits contingent on training and workflow integration.
  • Weak/limited evidence for high-complexity, fully agentic interventions (autonomous care robots, cross-sector case management by agentic systems) — large uncertainties in outcomes, safety, coordination, and scalability.

• Institutional determinants overshadow algorithms

  • Adoption and impact are driven by reimbursement models, regulatory clarity, data infrastructure and interoperability, workforce acceptance and training, liability frameworks, and equity/access policies.
  • Where institutional enablers exist (payment aligned to outcomes, clear certification pathways, integrated data), agentic AI deployments achieve fuller, measurable gains.

• Contributions from the review

• Agentic AI Framework (AAF 3.0): a sociotechnical taxonomy linking capability layers (task automation → multi-step coordination → autonomous decision-making), human-in-loop roles, institutional enablers, and evaluation metrics.
• Cross-domain synthesis formalising the inverse evidence–complexity relationship: empirical pattern that evidence strength E falls as intervention complexity C increases, moderated by institutional readiness R (conceptually E ∝ R / C).
• Phased sociotechnical roadmap: sequencing governance, reimbursement reform, and equity safeguards to move from pilot gains to safe, scalable adoption.

Data & Methods

• Review design: structured narrative review of 81 sources published 2020–2025, focusing on agentic AI applications within advanced-economy health and long-term care settings.
• Inclusion criteria: empirical or policy-relevant literature addressing autonomous or multi-step workflow AI systems in care delivery, workforce augmentation, care coordination, governance, or fiscal modeling; studies concerning advanced economies and ageing-related service demands.
• Evidence types: randomized and non-randomized evaluations, pilot implementations, qualitative implementation studies, modelling papers (microcosting, cost-effectiveness, limited macro simulations), and policy analyses.
• Synthesis approach: cross-domain mapping of interventions by complexity and evidence strength; identification of key mediating institutional factors; development of AAF 3.0 and a phased roadmap based on convergent findings. The review is narrative/structured rather than a quantitative meta-analysis due to heterogeneity in interventions, outcomes, and study designs.

Implications for AI Economics

• Short-term microeconomic effects

  • Likely: measurable per-unit productivity and efficiency gains in administrative and selected clinical/support tasks, potential reduction in marginal costs of certain services in LTC and home care.
  • Conditional: gains depend on reimbursement alignment (fee-for-service vs value-based), integration costs, retraining, and upfront capital investment.

• Uncertain macroeconomic and fiscal impact

  • No robust empirical evidence that agentic AI adoption will materially offset demographic-driven GDP growth slowdowns or ageing-related public expenditure increases at scale.
  • Macro effects depend on scale, diffusion, labor market adjustments (task reallocation versus displacement), wage responses in care sectors, and fiscal policy (taxation, transfer programs).

• Policy levers with high expected return

  • Reimbursement reform: move toward outcome- and value-based payments that reward cost-effective use of agentic AI and incentivize coordination across providers.
  • Governance sequencing: regulatory sandboxes, certification standards for agentic systems, staged approvals tied to risk tiers and monitoring obligations.
  • Institutional investments: interoperable data infrastructure, common APIs, privacy-preserving data governance, workforce training programs, and evaluation capacity.
  • Equity safeguards: subsidized access for low-income and rural populations, monitoring for algorithmic bias in care decisions, pathways to redress and accountability.

• Research and evaluation priorities for economists

  • Micro-to-macro modelling: integrate empirically grounded micro-level adoption scenarios into DSGE/CGE or sectoral simulation models to estimate fiscal and growth impacts under alternative policy regimes.
  • Longitudinal, system-level evaluations: RCTs or stepped-wedge designs that measure health outcomes, labor market effects, and cost offsets over medium-term horizons.
  • Distributional analyses: who benefits/loses across income, region, and care-worker demographics; incorporate equity constraints in cost–benefit assessments.
  • Implementation science: identify institutional preconditions that predict successful scaling, and formally test the roadmap interventions (governance + reimbursement + safeguards).

• Recommendations (actionable)

  • Prioritize institutional enablers before wide deployment: align reimbursement, legal frameworks, and data interoperability to amplify AI benefits.
  • Fund carefully designed pilots with built-in rigorous evaluation and scaling plans rather than assuming direct macrofiscal relief.
  • Use AAF 3.0 as a checklist for procurement and evaluation: match agentic capability level to organizational readiness and risk tolerance.
  • Policymakers and economists should be cautious in translating short-term productivity evidence into claims about macroeconomic fiscal moderation without explicit system-level modelling and scenario analysis.

In short: agentic AI can help mitigate certain operational pressures in ageing health systems, but its ability to resolve structural demographic and fiscal challenges is conditional on institutional reform, and large-scale macroeconomic benefits remain unproven.