The Invisible AI Income Shock: Why Job Losses May Understate AI’s Economic Impact on Social Care and Knowledge Work

At a small conference in 2026, a government official described the potential labour-market impact of artificial intelligence largely through a familiar equation: jobs would be lost, other jobs would be created, and if the eventual balance was favourable the economy should benefit. Sitting in the audience, however, was a much less tidy version of the same transition. Three people working in different forms of knowledge work had already experienced substantial reductions in paid work associated, at least in part, with AI. None had necessarily become unemployed. They simply had less work and less income.

That distinction may prove increasingly important. The economic effects of artificial intelligence will not necessarily arrive as millions of workers moving overnight from employment into unemployment. They could emerge much more quietly through fewer commissions, shorter projects, reduced billable hours, lower freelance rates, weaker graduate recruitment, slower replacement of departing employees, smaller administrative teams and organisations producing more output without proportionately increasing headcount.

For adult social care, this is not an abstract technology debate. Providers, local authorities, commissioners, consultancies, software suppliers and system organisations are already considering artificial intelligence and automation in care. The central question is therefore broader than whether AI will “take jobs”. It is whether AI begins changing the amount, type and economic value of human labour required to produce the services, administration, analysis and professional knowledge on which the sector depends.

If that happens, unemployment statistics alone may tell policymakers surprisingly little about the early stages of the transition.

The difference between losing a job and losing paid work

Traditional discussions of technological displacement often focus on employment status. A worker either has a job or does not. Yet contemporary labour markets contain employees, consultants, contractors, temporary workers, agency staff, freelancers, sole traders, portfolio workers and people whose earnings depend heavily on commissions, overtime or billable hours.

Artificial intelligence can affect each of those groups without creating a clear redundancy event.

A consultant may remain self-employed but sell 120 days of work rather than 180. A translator may retain established clients but receive substantially fewer assignments because routine material is translated automatically and only complex work is sent for human review. A writer may continue working but find that organisations increasingly create first drafts internally. A software developer may remain employed while the organisation decides that a smaller team can maintain the same development pipeline using AI coding tools.

In each case, labour has been displaced economically even though the individual can still answer “yes” when asked whether they are employed.

This distinction matters because the International Labour Organization's 2025 global assessment found that one in four workers worldwide was in an occupation with some exposure to generative AI, while emphasising that transformation rather than complete job replacement is the more likely outcome for most occupations. A further ILO analysis in 2026 warned that exposure indicators are not forecasts of redundancy and should not be interpreted as if every exposed occupation will disappear.

That caution is important. Exposure describes what AI may be capable of affecting. It does not reveal how organisations will redesign jobs, how rapidly adoption will occur, what customers will accept, how regulation will respond or whether falling production costs generate enough new demand to compensate for displaced work.

But transformation itself can have major economic consequences.

An income shock can exist without an unemployment shock

Consider a simplified economy containing ten professional workers earning £50,000 each. Together they receive £500,000 of labour income.

If AI-enabled productivity means all ten remain economically active but their average annual income falls to £40,000 through fewer hours, weaker rates, less overtime or reduced commissions, unemployment has not increased. Yet £100,000 of annual labour income has disappeared from those households.

The workers may reduce discretionary spending, postpone major purchases, contribute less tax and save less. Businesses receiving their expenditure then experience weaker demand. If similar changes occur across many occupations, the macroeconomic consequences can become meaningful even while the headline employment rate appears comparatively stable.

This is why workforce planning in an AI-enabled economy may increasingly need to examine more than establishment numbers and vacancy rates. Organisations may need to understand hours, spans of control, productivity per employee, changing occupational mixes, outsourced expenditure and the amount of previously purchased expertise that can now be produced internally.

The latest UK evidence does not establish that a large AI-driven income shock is already under way. Current labour-market conditions are affected by many forces including weak demand, interest rates, taxation, wage pressures, demographic change and wider economic uncertainty. Causation needs to be treated carefully.

Nevertheless, some early signals deserve attention.

UK evidence is beginning to show changes in how firms think about labour

In July 2026, the Bank of England reported what its regional Agents were hearing directly from businesses about AI adoption. Firms described productivity improvements from automating routine tasks and accelerating knowledge-intensive work, particularly in software development, finance, administration, customer service, professional services and content creation.

Importantly, the Bank reported that firms deploying AI effectively can increase output without a corresponding increase in employment and, in some cases, reduce staffing requirements. Some professional-services contacts were also reporting reduced graduate recruitment and weaker demand for administrative and junior employees.

That is not evidence of economy-wide mass unemployment. It is evidence of something more subtle: the relationship between output and labour demand may already be changing in some activities.

Separate analysis published by the UK Government's AI and the Future of Work Unit and LinkedIn in June 2026 found UK hiring overall was 14% lower year-on-year in April. Entry-level hiring was declining in 30 of 38 occupations examined. Among the sharpest falls were accountants, graphic designers and software engineers. The analysis explicitly cautioned that broader labour-market weakness and other factors mean these changes cannot simply be attributed to AI, but it noted that several of the fastest-declining occupations are also areas where AI capabilities have become particularly visible.

That matters because entry-level employment performs a function beyond producing today's output. Junior roles create tomorrow's experienced workers.

If AI performs document preparation, basic analysis, first-draft work, routine coding and administrative processing, organisations may logically ask why they should recruit the same number of junior employees. Yet senior professionals do not appear spontaneously. They develop through experience.

AI therefore creates a potential workforce resilience and continuity problem: a company can improve short-term efficiency while weakening the pipeline from which future expertise and leadership emerge.

The UK Government itself acknowledges how much remains unknown

It is important not to overstate current evidence. The Department for Science, Innovation and Technology and AI Security Institute published an assessment in January 2026 specifically examining AI capabilities and the UK labour market. Its conclusion was not that mass displacement was inevitable. Instead, it acknowledged that the available evidence still does not provide clear answers to many of the policy questions that matter most.

That uncertainty should encourage better measurement rather than complacency or alarmism.

The International Monetary Fund has estimated that around 40% of employment globally is exposed to AI, rising to around 60% in advanced economies. Its analysis distinguishes between work where AI may complement people and work where it could execute important tasks itself, potentially lowering labour demand, wages or hiring.

For the UK, the IMF's earlier analysis suggested particularly high occupational exposure because advanced service economies contain large numbers of cognitive and information-processing roles.

Exposure, however, is not destiny. A task becoming automatable does not mean that a business will automate it, that customers will accept automation, or that an employee becomes unnecessary. Adoption requires investment, organisational redesign, confidence, data, skills, governance and often substantial workforce adoption and digital capability.

This is where the debate moves from technological capability to organisational economics.

AI adoption is a change-management programme, not simply a software purchase

A model may demonstrate that a task can be completed more quickly, but organisations still have to decide how work should be redesigned around that capability.

A local authority introducing AI into commissioning analysis, for example, needs to consider information governance, data quality, professional accountability, assurance, procurement, workforce consultation, skills, bias, escalation and what decisions require human review. A provider automating quality reports or rota analysis needs to determine whether the information produced is sufficiently reliable to influence care delivery.

Employees may also have mixed incentives. A worker who believes a new system will improve their job may enthusiastically help design it. Someone who believes they are being asked to train the mechanism that eventually reduces their own hours may react very differently.

That does not make resistance irrational or employees anti-technology. It means leaders need to understand the human consequences of automation and workflow redesign.

The Digital Transformation Readiness Assessment is relevant here because genuine readiness involves considerably more than buying an AI licence. Organisations need governance, infrastructure, leadership, workforce capability, cyber resilience and a clear understanding of which processes technology should improve.

Poor implementation can actually increase workload. AI-generated material needs checking. Incorrect output creates rework. Automated customer-service systems can trap people in repetitive loops. Poorly designed interfaces move effort from one part of an organisation to another rather than removing it.

For this reason, technological potential and realised productivity should never be treated as the same thing.

Social care is an unusual test case because it simultaneously needs productivity and people

Adult social care makes the labour-market debate particularly interesting because the sector does not begin from a position of excess labour.

Skills for Care reported in June 2026 that England's adult social care vacancy rate had fallen to 6.2% in 2025/26, its lowest level for a decade. Yet this still represented around 96,000 vacant posts on any given day and a vacancy rate approximately three times that of the wider economy.

Longer-term demographic pressure remains. Previous Skills for Care projections have indicated that hundreds of thousands of additional posts could be required by 2040 if workforce growth follows increasing demand associated with an ageing population. The precise requirement will change as care models, technology, prevention and productivity evolve, but the underlying point remains: social care is trying to increase capacity in a constrained labour market.

This makes a simple narrative of “AI equals job destruction” particularly inappropriate for the sector.

If AI removes hours of duplicated administration, produces faster first drafts, identifies patterns in workforce data, improves scheduling, summarises meetings, supports quality monitoring and reduces repetitive data entry, it may allow existing workers to spend more time on activities that genuinely require human capability.

For frontline services, that could mean more time supporting people rather than completing avoidable administrative tasks. For managers, it could mean more time observing practice, coaching staff and solving operational problems. For commissioners, it could mean greater capacity for market engagement, provider oversight and strategic planning rather than repeatedly assembling information manually.

That is a potentially positive form of innovation and system-wide added value.

The economic question is what organisations then do with the released capacity.

Productivity gains can produce very different outcomes

Imagine an administrative team requiring 1,000 staff hours each month to complete a set of tasks. AI and workflow redesign reduce the requirement to 700 hours.

The organisation now has several choices.

• It can reduce staffing and retain the financial saving.
• It can maintain staffing and increase output.
• It can redeploy 300 hours into activities that previously lacked capacity.
• It can improve service quality while maintaining approximately the same cost.
• It can combine some staffing reduction with additional investment elsewhere.

The technology does not determine which outcome occurs. Management, funding, demand, regulation and organisational strategy do.

This distinction is central to understanding why AI can simultaneously represent a workforce threat and a solution to workforce scarcity.

In social care, there are compelling reasons to use productivity gains to release capacity rather than simply remove people. Many organisations operate with stretched management structures, demanding reporting requirements, recruitment pressures, high caseloads and limited improvement capacity. A technology that reduces low-value workload could strengthen rather than shrink the workforce if the released time is deliberately reinvested.

The risk is that financial pressure produces a different response. Providers operating on thin margins may reasonably use productivity improvements to reduce overheads. Commissioners facing constrained budgets may expect lower prices. Technology suppliers may promise savings that become embedded in future funding assumptions.

Over time, what began as optional productivity improvement can become a new expected cost base.

The invisible displacement may occur around care before it occurs within care

Direct personal care contains many tasks that remain difficult to automate because they involve physical presence, relationships, observation, dignity, trust, judgement and responsiveness to highly individual circumstances.

But adult social care exists inside a much wider professional ecosystem.

Providers purchase legal advice, recruitment services, marketing, consultancy, training, bid writing, translation, design, software development, policy support, research, auditing and administrative services. Commissioners and local authorities employ analysts, procurement professionals, project staff, communications teams and other knowledge workers.

Those surrounding activities may experience AI-driven productivity change faster than direct support.

A provider that previously commissioned an external consultant for a straightforward piece of research may now produce an initial analysis internally and buy only specialist review. A commissioner may use AI to compare tender documents before involving legal or procurement specialists. A communications team may generate routine material internally. A software business may use AI-assisted coding to increase the output of each developer.

Again, nobody necessarily becomes unemployed immediately.

But fewer professional hours are purchased.

This is why boards examining workforce risks and mitigation may eventually need to examine their external labour ecosystem as well as directly employed staff.

The change could be particularly significant for microbusinesses and self-employed specialists because their economic activity is poorly represented by simple headcount measures. A consultant whose workload falls from five days each week to three remains a consultant. A freelance writer whose annual commissions fall by a third remains self-employed. The income shock is real even though no redundancy appears in official statistics.

Reduced demand for junior work creates a second-order risk

There is another reason not to measure AI solely through current job losses.

Businesses can reduce labour demand by not creating jobs that would previously have existed.

This is much harder to observe than redundancy.

An organisation that would traditionally recruit four graduate analysts may recruit two because AI has increased the productivity of its existing team. Nobody receives a redundancy notice. Two jobs simply never appear.

Across an economy, this could gradually alter occupational entry routes.

The UK Government's June 2026 analysis of entry-level recruitment is therefore worth watching carefully. The current evidence does not establish AI as the cause of declining graduate or junior recruitment. The wider labour market has weakened. Yet the overlap between highly information-intensive work and rapidly improving AI capability makes continued monitoring important.

Within social care, similar questions apply to quality, commissioning and management pipelines. If junior administrative and analytical work becomes increasingly automated, organisations will need to consider how future managers, analysts and specialists acquire practical experience.

A strong continuous professional development model may eventually need to compensate for the disappearance of some traditional learning-by-doing tasks.

This is one reason AI strategy should not be separated from workforce strategy.

Productivity does not automatically translate into widely shared prosperity

The optimistic economic case for artificial intelligence is powerful.

Higher productivity means more output can be created from the same resources. Costs may fall. New services become economically viable. Businesses can expand. New industries can emerge. Workers equipped with AI can become substantially more productive. If those gains translate into higher real wages, lower prices, stronger investment and new demand, living standards can rise significantly.

This is why governments are understandably enthusiastic about AI as a potential source of economic growth.

But productivity and distribution are different questions.

If most productivity gains accrue to a relatively concentrated group of businesses, investors and highly augmented workers while large numbers of households experience weaker labour income, the economy can produce more while simultaneously creating pressure on mass purchasing power.

The IMF has highlighted this distributional risk. Its analysis notes that AI could increase income and wealth inequality depending on the balance between labour substitution, complementarity and returns to capital.

The outcome is therefore not determined simply by whether GDP rises.

Policymakers may need to understand:

• whether real median earnings are rising alongside productivity;
• whether working hours are changing by occupation;
• whether self-employed and freelance incomes are weakening;
• whether entry-level recruitment is contracting;
• whether labour's share of national income is changing;
• whether productivity gains are reducing prices;
• whether new demand is creating replacement employment; and
• whether gains are sufficiently distributed to sustain household consumption.

These measures would provide a much richer picture than unemployment alone.

Why the macroeconomic consequences could matter to social care

Adult social care is unusually exposed to the wider fiscal economy because so much provision depends directly or indirectly on public spending.

Local authorities purchase significant volumes of care. NHS bodies fund or contribute to some services. Providers depend on public-sector fee decisions, workforce policy and national funding settlements. People who fund their own care depend on household income, savings, pensions and asset values.

If AI eventually produces strong economy-wide growth, public finances could benefit considerably. Higher productivity can raise national income and expand the tax base, potentially creating more resources for health and social care.

But a transition characterised by weaker labour income and household demand could create a much more difficult picture even if technological productivity itself is impressive.

Lower household earnings can affect income-tax receipts, National Insurance revenues, consumption taxes and consumer demand. At the same time, governments may face pressure to support displaced or lower-income workers, invest in retraining and respond to regional or occupational disruption.

Social care leaders therefore have an interest in the wider distribution of AI's economic gains. The sector's sustainability cannot be separated entirely from national growth, tax revenue, household wealth and public expenditure.

Headline unemployment may be a lagging indicator

None of this means unemployment is irrelevant. Significant occupational displacement could eventually produce higher unemployment if new demand and new jobs do not emerge quickly enough.

The argument is narrower and more important: unemployment may be one of the later indicators rather than the earliest.

Before redundancies appear at scale, organisations may freeze vacancies, reduce agency use, stop replacing leavers, cut freelance budgets, consolidate roles, lower recruitment at entry level and expect AI-enabled staff to absorb additional activity.

The Bank of England's 2026 business intelligence already describes some firms achieving staffing reductions through natural attrition rather than conventional job-cutting programmes, while organisations across sectors remain focused on efficiency and technology.

A labour market can therefore adjust through thousands of small organisational decisions rather than one dramatic wave of redundancy announcements.

For providers, commissioners and boards, the Predictive Workforce Risk Module illustrates the broader principle that workforce risk needs to be understood through multiple indicators. Vacancy rates matter, but so do turnover, capacity, continuity, role dependencies and changes in how labour is deployed.

AI could also create demand that does not currently exist

There is an important counterargument to the invisible-income-shock thesis.

When technology dramatically reduces the cost of producing something, society often consumes much more of it.

A person who would never pay hundreds of pounds for conventional professional support may pay a small subscription for an AI-enabled service. A provider that could not afford continuous analytical support may use an inexpensive AI tool every day. A commissioner may be able to interrogate data far more frequently than was previously practical.

This can create new markets rather than simply redistribute existing work.

Workers can also become dramatically more productive without suffering lower income. An experienced professional who uses AI to complete routine research and first drafts may serve more clients, undertake higher-value work and increase earnings. A care provider may expand without needing administrative costs to rise proportionately. A technology business may use AI productivity to build products that would previously have been uneconomic.

The eventual labour-market outcome will therefore reflect at least three forces:

• substitution — AI performs work previously purchased from people;
• augmentation — people become more productive while remaining central to the work; and
• demand creation — lower costs and new capabilities create additional services, businesses and occupations.

The balance between those effects is unknown.

That is why confident predictions of either a job apocalypse or an uncomplicated productivity boom should be treated cautiously.

The pace of transition may matter as much as the eventual destination

Previous technological revolutions have repeatedly changed occupations and created new ones. The long-run capacity of economies to generate new human wants and new forms of employment should not be underestimated.

Artificial intelligence nevertheless has characteristics that could make the transition unusually rapid.

Much of the necessary digital infrastructure already exists. New model capabilities can be distributed through software people already use. Organisations do not necessarily need to build factories, install physical machinery or wait years for capital projects before experimenting with AI-enabled knowledge work.

Economic institutions move more slowly.

Education programmes take years to redesign. Employment structures require consultation and change management. Public-sector procurement can be lengthy. Tax systems develop incrementally. Regulation follows evidence. Professional standards need to adapt. People need time to acquire different skills.

The more rapidly technological capability changes relative to institutional adaptation, the greater the risk of a difficult transition even if the eventual long-term outcome is positive.

This timing issue may be particularly important for governance and leadership. Organisations need enough ambition to capture productivity improvements without moving faster than their ability to manage safety, workforce consequences, information risk and accountability.

The Governance Maturity Assessment reflects the importance of maintaining that wider oversight. AI adoption should become a board-level organisational question where the consequences are material, rather than remaining solely an IT project.

Social care should distinguish human work from unnecessary human workload

One of the most constructive approaches for adult social care would be to avoid framing the issue as a choice between protecting every existing task and automating as much as possible.

Those are both poor objectives.

The sector should instead ask which activities genuinely benefit from human attention.

Supporting someone through distress, recognising subtle deterioration, negotiating risk with a person and their family, providing intimate personal care with dignity, resolving complex safeguarding concerns, leading a team through crisis and making accountable decisions under uncertainty remain profoundly human activities.

Copying information between systems, repeatedly restructuring the same data, creating routine first drafts, manually searching large policy collections and performing predictable administrative transformations are different.

If AI can safely reduce the second group, protecting them merely because they currently occupy paid hours would be difficult to justify in a sector with scarce resources and substantial unmet demand.

The objective should be better use of human capability, not maximum preservation of existing workflows.

That connects AI directly to safe staffing and deployment. Productivity should ultimately be judged by whether organisations can deploy appropriate capability where people receiving support need it most.

But organisations should understand who bears the economic saving

Productivity discussions can become incomplete when they identify a saving without asking where it goes.

Suppose an AI implementation saves a provider £200,000 each year in administrative labour. Several outcomes are possible.

The provider could improve wages. It could increase margins and strengthen financial resilience. Commissioners could ultimately capture some of the saving through lower prices. The organisation could invest in frontline capacity, technology or quality improvement. Shareholders or owners could receive higher returns. Employees could work fewer hours while maintaining income. Some roles could disappear.

Each represents a different distribution of exactly the same productivity improvement.

For organisations making significant workforce changes, fair work and responsible employment therefore becomes relevant to AI governance. There will be legitimate commercial decisions about efficiency, but the transition is more likely to retain trust where organisations explain why technology is being introduced, what work will change and how employees will be supported.

The wider economy faces essentially the same question at much larger scale: if AI produces enormous gains, how are those gains transmitted into household living standards?

Government may eventually need a wider AI labour-market dashboard

If policymakers want early warning of harmful labour-market effects, unemployment should sit inside a broader set of measures.

ONS data already allows government and researchers to examine earnings, employment, vacancies and working hours. Average regular earnings in Great Britain were growing by 3.5% annually in April to June 2026, but only 0.5% in real terms after CPIH inflation. These aggregate figures do not isolate AI effects, but they illustrate why earnings need to be examined alongside employment.

A useful AI-transition dashboard could increasingly examine:

• real earnings by occupation and income group;
• hours worked and involuntary reductions in hours;
• self-employed earnings and business income;
• freelance and contract volumes where data can be developed;
• vacancy creation and vacancy withdrawal;
• entry-level and graduate recruitment;
• occupational transitions;
• redundancy and natural attrition;
• productivity by sector;
• business investment in AI and automation;
• labour share of national income;
• household consumption; and
• the creation and growth of occupations associated with AI.

None should be interpreted alone.

A fall in junior hiring could reflect recession rather than AI. Falling freelance rates could result from global competition. Higher productivity could come from ordinary software investment. Labour-market research will need credible counterfactuals rather than attributing every change after 2022 to generative AI.

But measurement should be designed around the mechanisms through which disruption could occur.

Providers and commissioners should monitor their own microeconomics

National statistics will inevitably lag behind organisational change. Social care leaders can therefore learn something by examining their own operating models.

A commissioner introducing AI-assisted analysis might measure not only whether reports are produced faster, but whether external consultancy expenditure changes, how analyst workloads change and whether more strategic work is actually completed.

A provider introducing automated quality reporting could assess management hours before and after implementation, whether administrative staffing changes, how quickly concerns are identified and whether Registered Managers gain meaningful additional time with teams.

A care-technology company could examine output per developer alongside junior recruitment and future skills requirements.

The Digital Twin Scenario Modeller demonstrates the value of considering workforce, capacity and service consequences together. Organisations contemplating substantial AI-enabled redesign should model more than immediate cost reduction: they should consider what different staffing structures mean for resilience, capability and future demand.

Four scenarios illustrate how differently the same technology can affect labour

Scenario one: the consultancy budget contracts

A provider previously purchases 80 days of specialist external support each year. Managers begin using AI to conduct routine research and prepare initial drafts. The provider still values human expertise but now purchases 45 days focused on review, complex analysis and assurance.

No consultant has necessarily become unemployed. But 35 days of professional labour demand have disappeared from that organisation.

Scenario two: the commissioner expands capacity

A commissioning team spends large amounts of time manually summarising provider reports. An AI-enabled workflow reduces that work substantially. The authority does not reduce headcount. Instead, commissioners spend the released capacity visiting services, analysing markets and addressing provider failure earlier.

AI has displaced tasks but increased the value produced by the same workforce.

Scenario three: the provider does not replace a vacancy

A quality administrator leaves. During the recruitment period, the provider discovers that new automation allows the remaining team to manage the workload safely. The vacancy is withdrawn.

Nobody has been made redundant. Yet one job that would probably have existed no longer does.

Scenario four: frontline capacity increases

A homecare organisation automates significant parts of scheduling administration, reporting and routine communication. Savings are partly reinvested in care-worker pay and additional supervisory capacity. Retention improves and managers spend more time with frontline teams.

In this scenario AI has reduced some administrative labour demand while strengthening the care workforce.

These scenarios demonstrate why a single national figure for “jobs created minus jobs destroyed” cannot capture the whole transition.

The question for social care is not whether to resist AI

Adult social care has too many pressures to reject useful productivity improvements simply because existing patterns of work may change.

At the same time, the sector should resist simplistic claims that technology automatically creates efficiency, solves workforce shortages or can replace professional judgement.

Strong adoption requires evidence.

Leaders should understand what task is being changed, what measurable benefit results, what new risks appear, what happens to the workforce, whether the experience of people receiving support improves and whether apparent cost reductions merely transfer workload somewhere else.

That is the difference between digital enthusiasm and disciplined transformation.

The bigger economic question may be how much human labour society needs to purchase

Artificial intelligence could ultimately prove less disruptive than its most enthusiastic advocates and most alarmed critics predict. Organisational adoption may remain slow. Human preferences may sustain demand for human-delivered services. New occupations may absorb displaced workers. Falling costs may create enormous new markets.

But another possibility deserves serious policy attention.

AI may allow many organisations to purchase materially fewer hours of human cognitive labour while continuing to produce the same or greater output.

If that occurs across enough occupations, the initial consequence may not be mass unemployment. It may be millions of individually modest decisions: one vacancy not replaced, one less freelancer commissioned, two fewer graduate recruits, a shorter project, a smaller agency contract, reduced overtime, a narrower support function.

Each decision is economically rational in isolation.

Together they could represent a significant shift in the relationship between productivity, employment and household income.

What should we watch next?

The coming years should therefore be judged using more than spectacular demonstrations of AI capability and headline predictions of future employment.

We should watch whether organisations genuinely increase productivity. We should watch whether new demand absorbs the capacity released. We should watch earnings and hours as closely as unemployment. We should examine entry-level recruitment because today's junior roles create tomorrow's experienced workforce. We should understand whether productivity gains reach workers and consumers or become increasingly concentrated.

And within social care, we should ask a particularly practical question:

Does AI enable scarce human capability to move towards the work where people need humans most?

If it does, the technology could become an important part of solving one of adult social care's most persistent problems: providing more and better support despite severe workforce and financial constraints.

If instead AI weakens incomes and labour demand across the surrounding economy faster than new opportunities emerge, the consequences could eventually affect the tax base, public spending, household demand and therefore the financial environment in which social care itself operates.

Neither outcome is predetermined.

The responsible position is therefore neither resistance nor blind optimism. It is to embrace useful capability, measure the consequences carefully and recognise that the economic transition may appear in places that conventional employment statistics do not immediately reveal.

The defining AI labour-market question may turn out not to be simply “How many jobs disappeared?”

It may be:

“How much paid human work did the economy still require, who benefited from the productivity that replaced it, and what happened to the people whose incomes depended on selling that work?”