AI is changing the nature of work by automating tasks, not entire occupations: about 9% of jobs are fully automatable, yet almost every role will be affected. Who wins or loses will hinge on firms' training choices and public policy, with displacement concentrated among older and lower-mobility workers.

Main Finding

AI primarily automates tasks, not whole jobs. Only about 9% of jobs are fully automatable today, but nearly every job is being affected as AI changes the task mix inside occupations. How individuals, firms and policymakers respond to task-level change — especially by investing in adaptation and training — will determine who benefits and who loses.

Key Points

• Core distinction: tasks vs jobs. Thinking in tasks exposes much more precise policy and firm responses than the blunt “robots taking jobs” narrative.
• Quantitative headline: ~9% of jobs are fully automatable, yet virtually all occupations contain tasks that AI can augment or replace.
• Five lessons from past technological transitions:
• Fear and hype are temporary; long-run impacts depend on adaptation.
• Who adapts determines who wins: winners are those who re-skill and redeploy capital effectively.
• Firms that invest in worker training outperform those that cut labor.
• Transitions are uneven across sectors, regions and demographic groups.
• New tasks and jobs emerge alongside displacement, but not in the same places or for the same people.
• Three categories of skills:
• Skills AI is absorbing (routine, information-processing, some pattern-recognition tasks).
• Skills needed to work alongside AI today (prompting, oversight, data curation, human-in-the-loop management).
• Skills likely to remain human-anchored longer term (complex judgment, social coordination, creativity, contextual problem-solving).
• Four channels by which AI affects labor markets:
• Evolution of existing roles (task reallocation within jobs).
• Creation of entirely new roles (AI trainers, prompt engineers, AI compliance officers).
• Redistribution across geographies and demographics (remote work, offshoring patterns, local agglomeration effects).
• Selective displacement concentrated among older and lower-mobility workers who face higher frictions in switching tasks or locations.

Data & Methods

• Evidence sources: aggregated labour-market statistics, task-level occupation decompositions, historical analogies from prior technology waves, and emerging workplace case studies/field evidence.
• Measurement approach:
  • Task-based analysis: jobs decomposed into constituent tasks; each task assessed for susceptibility to current AI capabilities.
  • Classification of “fully automatable” jobs requires that all or almost all core tasks are automatable — explaining the low (~9%) figure.
  • Complementary use of workplace evidence to observe real-world task reallocation, hiring patterns, and firm training behavior.
• Strengths: more granular, policy-relevant view than occupation-level automation estimates; triangulates quantitative assessment with real-work evidence.
• Limitations and uncertainties:
  • Rapidly improving capabilities may change the set of automatable tasks.
  • Estimates depend on task definitions and assumptions about adoption speed and cost.
  • Distributional impacts depend on institutional responses that are hard to predict.

Implications for AI Economics

• Measurement and modeling:
  • Move from occupation-level to task-level metrics in empirical work and macro/labor models.
  • Incorporate heterogeneity in worker mobility, firm training behavior, and adoption lags.
  • Study complementarities between AI and human skills (which tasks are augmented vs substituted).
• Policy design:
  • Prioritize active labour-market policies: subsidized retraining, on-the-job training incentives, and portable credentials that map to task bundles.
  • Encourage employer-led upskilling (tax credits, co-funded training) because firms that train perform better and mitigate displacement.
  • Strengthen safety nets and transition assistance focused on older and low-mobility workers who face the highest frictions.
  • Regionally targeted policies to address geographic redistribution (infrastructure, local training, remote-work facilitation).
• Firm strategy:
  • Firms should audit task composition of roles to identify augmentation opportunities, redeploy workers into higher-value tasks, and invest in human+AI workflows.
  • Hiring and compensation practices should reflect changing task demands (reward supervision, curation and coordination skills).
• Distributional and welfare concerns:
  • Expect uneven gains: workers who develop complementary skills or are in adaptable firms will capture most benefits; others risk displacement.
  • Research should quantify long-run wage and employment dynamics across cohorts, sectors and regions to inform redistributive policy.
• Research priorities:
  • High-quality panel data on tasks, on-the-job activity, training investments and firm adoption decisions.
  • Field experiments evaluating training programs, employer incentives, and AI rollout governance.
  • Models linking micro task reallocation to aggregate productivity and inequality outcomes.

Bottom line: the right analytical frame is task-centered. That reframing leads to more actionable policies and firm strategies — emphasize adaptation, targeted training, and institutions that lower frictions for workers transitioning into AI-complementary tasks.