AI, Jobs and the Future of Work: Who Gains Productivity, and Who Bears the Cost?

The AI Funding Files: Following the Money, Examining the Impact

Will artificial intelligence create better jobs, reduce workloads and help businesses grow — or will it make some workers easier to replace?

The answer is not as simple as the headlines suggest. AI can automate certain tasks without eliminating an entire job. It can help a worker complete work more quickly, but it can also change what employers expect, which skills they value and how many people they need.

The UK Government's January 2026 assessment of AI and the labour market found evidence of productivity gains in some tasks, alongside early signs that job advertising had fallen more sharply in occupations more exposed to AI. However, the report cautioned that the evidence does not establish that AI alone caused those changes.

Meanwhile, the Office for National Statistics reported in July 2026 that businesses were more likely to integrate AI skills by training existing staff than by replacing workers. Around 10% of businesses surveyed reported automating or replacing roles with AI technologies, while widespread changes in total workforce numbers had not yet emerged.

The real issue is not just whether AI increases productivity. It is who benefits from that productivity, who has a say in how work changes, and what happens to the people whose jobs or opportunities are affected.

1. What does AI actually change about work?

Most jobs consist of multiple tasks. A teacher may prepare resources, explain ideas, assess work and support pupils. A care worker may complete records, organise activities, communicate with families and provide direct support. A musician may write lyrics, compose, rehearse, record, promote and negotiate contracts.

AI may be able to assist with some tasks while having little ability to perform others safely or effectively. It might draft a document or generate an initial musical idea, but that does not mean it can replace the judgement, relationships, practical skills and accountability required to do the whole job.

This distinction matters. A job being exposed to AI does not automatically mean it will disappear. Exposure means that some of its tasks could potentially be affected by the technology. The actual outcome depends on how employers deploy AI, how workers adapt, the quality of the systems and whether human involvement remains necessary.

For some workers, AI could reduce repetitive tasks and create more time for complex work. For others, it could increase monitoring, raise workloads or reduce opportunities to enter a profession.

2. What does the evidence say about UK jobs?

The Department for Science, Innovation and Technology and the AI Security Institute published an assessment of AI capabilities and the UK labour market on 28 January 2026.

The report brought together evidence on the capabilities of AI systems, which tasks and occupations may be affected, productivity research and early changes in recruitment. It found that AI can improve the speed of certain tasks, but that results vary by occupation and setting.

Analysis of UK job postings found an association between greater occupational exposure to AI and a decline in advertised vacancies. One analysis estimated that a one-standard-deviation increase in AI exposure was associated with a 3.9% reduction in posting volume, with the decline becoming statistically significant around seven months after ChatGPT's release and returning to previous levels after approximately 20 months.

This is an association, not proof that AI caused the change. Recruitment can also be affected by interest rates, economic conditions, business restructuring, changes in demand and other technologies.

The report also found that some studies measured productivity improvements in specific tasks, including writing and software development. These results should not be treated as a guarantee that every worker or business will achieve the same gains.

Source: DSIT and the AI Security Institute, Assessment of AI Capabilities and the Impact on the UK Labour Market, January 2026.

3. Are businesses replacing workers with AI?

Current evidence suggests a mixed picture rather than a simple story of mass replacement.

The Office for National Statistics' July 2026 analysis of AI in UK businesses found that, among businesses with 10 or more employees, reported AI use had risen from around 12% in late 2023 to around 35% by June 2026.

Among businesses adopting AI, training or retraining existing staff was the most commonly reported approach to integrating AI-related skills. Around 10% reported automating or replacing roles with AI technologies. The ONS also found that only 11% of businesses reported that more than half of their workforce had received AI-related training.

These figures suggest that many businesses are adapting existing work rather than immediately restructuring their entire workforce. However, the effects may differ between sectors, companies and individual occupations, and national averages can conceal changes affecting particular groups.

There is also a difference between replacing an employee and reducing future recruitment. A business might keep its current staff but hire fewer people as employees leave. The effect may therefore appear in entry-level vacancies before it becomes visible in total employment figures.

Source: Office for National Statistics, Artificial Intelligence in UK Businesses: 2023 to 2026.

4. Who gains when AI makes work faster?

Suppose an employee uses AI to complete a task in half the time it previously required. That may be a genuine productivity improvement. But what happens next depends on the employer's decisions.

The employee might use the saved time to provide better service, learn new skills or complete work that previously had to be postponed. The employer might increase output, reduce operating costs, improve profits or invest in new services. In some circumstances, the business may need fewer staff to perform the same amount of work.

Productivity gains can therefore be distributed in different ways. They may lead to higher wages, improved services, shorter working hours, increased profits or reduced employment. These outcomes are not automatic consequences of the technology itself.

For workers to benefit, the way AI is introduced matters. Training, fair workplace policies, employee consultation and a clear plan for using the time saved can influence whether the technology improves working life or simply increases the amount of work expected.

Policymakers should also examine how gains are distributed between companies, investors, workers, customers and the wider public. An increase in productivity is an economic measure; it does not, by itself, tell us whether people's lives have improved.

5. Which jobs and tasks may be affected?

AI can affect work across many sectors, but exposure varies according to the tasks involved and the capabilities of the technology.

• Administration and clerical work: drafting routine correspondence, organising documents, summarising information and processing standard enquiries.
• Customer services: answering common questions, assisting with bookings and routing enquiries, while more complex cases may still require human judgement.
• Professional services: assisting with research, document review, coding, analysis and the preparation of initial drafts.
• Education: helping educators prepare resources and adapt materials, while teaching, relationships and safeguarding remain important human responsibilities.
• Healthcare: supporting administrative work, research and selected clinical tasks under appropriate professional oversight.
• Creative industries: generating draft images, music, text and video, potentially changing production costs, working methods and demand for certain services.
• Industrial work: supporting quality control, planning, robotics and other applications that combine software with physical machinery.

This list describes areas where tasks may change; it is not a prediction that these occupations will disappear. In many settings, the outcome will depend on how well AI can perform the task, what errors cost, what customers expect and whether human skills remain essential.

6. The entry-level jobs question

One potential risk is that businesses may use AI to automate tasks traditionally assigned to junior employees. These tasks can include preparing first drafts, basic research, routine coding, document checking and administrative support.

Some of this work may be repetitive, but it can also be how people learn a profession. Junior workers gain experience by completing simpler tasks, receiving feedback and gradually taking on more responsibility.

If employers automate too much of this work without creating alternative training opportunities, new entrants may find it harder to develop the skills and experience needed for more advanced roles.

This could create a long-term skills problem. Organisations may reduce entry-level recruitment today and later discover that they have too few experienced people who understand the work, can evaluate AI outputs or manage complex cases.

Employers should consider how apprenticeships, graduate schemes, work placements and supervised training will change. AI can support learning, but it should not become an excuse to remove the pathways through which people acquire practical experience.

7. Who has access to AI skills and training?

AI adoption is not uniform across the UK. Businesses differ in their size, resources, infrastructure, sector and ability to provide training.

The Government's UK Business Data Survey 2026 found that businesses handling digitised data reported different rates of AI use according to size. Larger businesses were more likely to use AI than microbusinesses and sole traders, while adoption also varied considerably by industry and region.

Skills England's 2026 annual skills report highlights gaps in AI knowledge among employers and workers. It argues that many people will need practical AI literacy — the ability to use tools, check outputs and integrate them responsibly — even if they do not become specialist AI developers.

Without accessible training, people who already have strong digital skills and access to professional development may be better placed to benefit. Others may be expected to use new systems without sufficient preparation or may struggle to access the jobs being created.

A fair transition should provide training during paid working hours where appropriate, accessible learning materials, practical support for small employers and opportunities for people changing careers.

Sources: UK Business Data Survey 2026; Skills England Annual Skills Report 2026.

8. What does this mean for disabled and neurodivergent workers?

AI can create opportunities for disabled and neurodivergent people, including tools for speech-to-text, text-to-speech, communication, planning, summarisation and creative production. Used well, these technologies may help people express ideas, manage certain tasks and participate in work in ways that suit their needs.

But access to a tool does not automatically produce inclusion. Some systems may be difficult to use with assistive technology, misunderstand alternative communication styles or create new barriers for people who need more time, human support or accessible information.

There are also risks if employers use AI to screen job applicants, assess performance, monitor productivity or infer characteristics about workers. Automated systems may reproduce bias in their data or disadvantage people whose communication, behaviour or working patterns differ from the assumptions built into the system.

Employers should assess accessibility and equality impacts before using AI in recruitment or employment decisions. They should provide reasonable adjustments where required, explain how AI is being used and ensure that workers have a meaningful way to question inaccurate assessments.

Disabled and neurodivergent people should also be included in designing and testing workplace AI. Their experiences can help identify barriers that developers and employers might otherwise overlook.

9. Creative work, music and AI

AI tools can help musicians, writers, filmmakers, designers and other creatives explore ideas, produce drafts, experiment with styles and carry out technical tasks. Independent creators may find that tools which were previously expensive or difficult to access are becoming more available.

However, the technology can also change the market for creative labour. Some clients may commission fewer paid drafts, use AI-generated content instead of hiring a specialist, or expect artists to produce more work for the same fee.

There are additional questions about copyright, training data, permission, voice and likeness. A creator's ability to use an AI tool does not settle whether the material used to train it was authorised or whether a generated output infringes someone else's rights.

Musicians and other creators should understand the terms of the tools they use, keep records of their original work and obtain appropriate permission when using another person's voice, likeness or protected material.

For disabled and neurodivergent creatives, AI may be a useful assistant that expands access to production and storytelling. It should not be used to dismiss their original work, assume that technology can speak for them or remove their control over how their identity and creative contributions are represented.

10. Does AI improve productivity?

Productivity describes how much output is produced from a given amount of input, such as working time or resources. AI may improve productivity when it helps people complete useful work more effectively.

The Government's January 2026 labour-market assessment reported productivity gains in some tasks, including writing and software development. However, it also noted that estimates differ between studies and settings.

There is an important difference between completing one task faster and improving the productivity of an entire organisation. Time saved on drafting may be offset by checking errors, correcting poor outputs, integrating systems, training staff or managing new risks.

Productivity also depends on the quality of the output. A system that generates a report quickly but includes incorrect information may create more work rather than less. In healthcare, education, safeguarding or other sensitive settings, the cost of an error may be far greater than the time saved.

Businesses should evaluate AI against a realistic baseline. They should measure the total time and cost involved, the quality of the results, the frequency of errors and the effects on workers and customers.

11. Could AI increase inequality?

AI may increase inequality if the benefits are concentrated among people and organisations that already have access to capital, computing resources, specialist expertise and professional networks.

Large employers may be better placed to purchase tools, integrate them into existing systems and provide staff training. Smaller organisations may face greater costs relative to their resources, while individual workers may have limited influence over how their employers introduce the technology.

There may also be differences between people who own or invest in technology and those whose work is changed by it. If productivity gains increase profits but wages and job quality do not improve, the economic benefits may not be shared evenly.

These outcomes are not inevitable. Public policy, education, competition, employment protections, accessible training and responsible business practices can influence how the benefits are distributed.

The question for policymakers is not simply how to accelerate adoption. It is how to ensure that workers, smaller businesses and communities have a fair opportunity to benefit from the transition.

12. What should employers do?

Employers should treat AI adoption as a workplace change that requires planning, communication and evaluation — not simply a software purchase.

• Explain the purpose: tell workers which tasks the system will support and why it is being introduced.
• Consult employees: involve the people who understand the work and will be affected by changes.
• Provide training: help staff use AI safely, verify outputs and understand limitations.
• Protect job quality: consider workload, monitoring, working hours and whether productivity gains can improve working conditions.
• Maintain human judgement: do not rely on automated outputs for consequential decisions without appropriate review.
• Assess equality and accessibility: check that tools do not create avoidable barriers or unfairly disadvantage workers.
• Be transparent about performance: evaluate errors, time savings, costs and effects on customers and staff.
• Plan for change: offer retraining and support if roles change, and consider how new entry-level opportunities will be maintained.

Employers should also be careful about entering confidential company information, employee records or personal data into AI tools that have not been approved for that purpose.

13. What should government measure?

Public investment and policy should be assessed by more than the number of businesses adopting AI or the amount of money spent on training.

A useful accountability framework should track:

• Employment, recruitment and redundancy trends across occupations and sectors.
• Changes in pay, working hours, job quality and employment security.
• Whether productivity gains translate into better services, higher wages or reduced costs.
• Access to training and whether it leads to recognised skills or improved employment opportunities.
• Effects on apprenticeships, graduate recruitment and other entry-level pathways.
• Differences in outcomes by region, business size and relevant equality groups.
• Accessibility and the experiences of disabled and neurodivergent workers.
• The costs of implementation, errors and ongoing maintenance.
• Whether workers are consulted and can challenge unfair or inaccurate automated decisions.

Evaluation should distinguish between changes that occur alongside AI adoption and changes that can reasonably be attributed to it. Economic conditions, business strategy and other technologies can also affect employment.

14. Questions workers should be asking

• What tasks will AI change in my role, and what will remain my responsibility?
• Will I receive training and time to learn how to use the system?
• How will my employer assess the quality of AI-assisted work?
• Will productivity gains lead to better working conditions or simply higher targets?
• Is AI being used to monitor, rank or assess my performance?
• Can I challenge an inaccurate automated assessment?
• How will my employer protect my personal information?
• What happens to entry-level jobs and training opportunities?
• Are accessibility and reasonable adjustments being considered?
• If my role changes significantly, what support will be available?

Conclusion: productivity should benefit people, not just systems

AI is changing how some work is organised, and there is evidence that it can improve the speed of particular tasks. There are also early indications that recruitment and workplace roles may change in occupations more exposed to the technology.

But the scale and direction of the long-term effects remain uncertain. AI exposure does not automatically mean job loss, and reported productivity improvements do not automatically mean better wages, working conditions or public services.

The outcome will depend on choices made by employers, workers, policymakers, educators and technology developers. Training, consultation, fair employment practices and accessible design can help ensure that people are not left to manage the transition alone.

The real measure of AI progress should not be how much work a machine can take over. It should be whether the technology helps people do meaningful work, creates fair opportunities and shares the benefits of productivity more widely.

The AI Funding Files examines public commitments, published evidence and the questions that should be asked as AI policy develops. Early evidence should not be treated as a definitive forecast of future employment.