AI and the Labour Market, Signal, Noise and Structural Change

Narrative

VC and analyst coverage treats generative AI primarily as a work-redesign and enterprise-productivity thesis, not as demonstrated aggregate labour-market displacement. McKinsey Global Institute’s 2023 US model and 2024 Europe-US update estimate that 30% of current hours could be automated by 2030 in the United States, with transitions concentrated among lower-paid office, production and customer-service work. These are scenario models of task exposure and adoption, explicitly entangled with ageing, pandemic-era occupational churn, infrastructure investment, net-zero investment and e-commerce, rather than observed causal estimates of post-2022 AI job losses.

The more recent Gartner material offers a partial corrective to early headcount rhetoric. Its 2025 supply-chain survey found weekly time savings for desk workers but much smaller team-level savings, while its December 2025 analysis attributed less than 1% of announced first-half 2025 layoffs to AI productivity gains and argued that most reported cuts reflected wider strategic repositioning. Gartner’s 2026 workforce-cost thesis also warns that removing junior roles can weaken talent pipelines and create costly rehiring, but its claims are forecasts and proprietary client research rather than administrative employment evidence.

Sequoia and CB Insights describe a movement from copilots to agents and service-delivery "autopilots", with software engineering the leading use case. Their evidence is mainly venture-market mapping, product adoption and investment activity: CB Insights reports that 349 S&P 500 firms had documented external AI activity in 2023-25, but five firms accounted for 32% of that activity and the median active firm had only three to four documented AI partnerships. This concentration, together with Bain’s finding that 87% of surveyed companies were developing, piloting or deploying genAI by early 2024, indicates broad experimentation but does not establish widespread realised productivity or employment effects.

Forrester’s reports consistently predict that more jobs will be reshaped than eliminated, and its 2025 software-work analysis identifies a specific risk to entry-level training as routine artefact-production tasks are automated. Its forecasts and interviews support work-redesign hypotheses, particularly in software, but should not be read as evidence of realised changes in vacancies, wages or employment. Across the lane, the strongest disconfirming theme is that individual time savings do not automatically translate into team output, payroll reduction or economy-wide productivity, so official ONS, BLS, Eurostat, OECD, ILO and IMF series remain necessary to test these theses.