AI Can Empower You or Overwhelm You
Artificial intelligence can save somebody ten minutes writing an email, or allow the same person to attempt work they could not previously have done at all. The distinction between those two uses may prove more important than the headline numbers showing how quickly AI is being adopted.
The anxiety surrounding the technology is understandable. If a machine can write, calculate, translate, analyse, code, research and create in seconds, work that once required specialist knowledge, considerable time or another person suddenly looks vulnerable. Some occupations will shrink, some tasks will disappear and skills that once commanded a premium will become inexpensive when software can reproduce them almost instantly.
There are already signs of pressure at the beginning of some careers. Research by Stanford’s Digital Economy Lab, using payroll data covering millions of US workers, found that employment among 22 to 25-year-olds in highly AI-exposed occupations stood 19% below where it would have been had it kept pace with less-exposed peers by mid-2026. The researchers found no comparable gap among experienced workers and no evidence of widespread economy-wide displacement.
Concentrating entirely on what artificial intelligence might take away, however, overlooks something already happening on the other side of the equation. AI is giving individuals access to capabilities that previously required more time, more money, more education or another person. The emerging divide may run between people who use AI largely for convenience and those who learn how to extend themselves with it. Access to the technology is rapidly becoming commonplace; extracting meaningful capability from it is another matter.
Briefing
- ChatGPT now has more than one billion weekly active users, according to OpenAI’s August 2026 data.
- OpenAI found that 43.5% of occupation-specific work messages involved tasks associated with an occupation other than the user’s own.
- A study of 5,179 customer-support agents found AI assistance increased productivity by 14% on average, with much larger gains among novice and lower-skilled workers.
- METR found experienced open-source developers using early-2025 AI tools took 19% longer to complete tasks despite believing the technology had made them faster.
- Danish labour-market research found widespread chatbot adoption and reported time savings but no detectable effect on earnings or recorded hours within two years.
Expanding Individual Capability
For most of the industrial era, expanding what an individual could do usually meant acquiring another skill or involving another person. An engineer who needed legal advice consulted a lawyer. A small company wanting market research paid a specialist. Someone who needed software either learned programming or employed a developer. A student struggling with mathematics found a teacher or worked through a textbook. A business entering another country needed translators, researchers and local expertise. Those professions remain necessary, but the barrier to performing useful work outside one’s own speciality is falling rapidly.
OpenAI has attempted to measure this in an analysis of more than 800,000 messages from US ChatGPT users. It found that 16.8% of work-related messages involved activities associated with another occupation. Once generic activities such as writing, summarising and scheduling were removed, 43.5% of occupation-specific messages crossed occupational boundaries. Among designers the figure was 75%, while customer-experience workers reached 77%, human-resources workers 69%, legal workers 56% and marketers 53%.
Those figures need some caution. OpenAI developed the classification system, and a message asking an AI system to perform work associated with another occupation does not demonstrate that the resulting work was correct, useful or commercially viable. What the data does show is that people are attempting tasks beyond the conventional boundaries of their own occupations on a substantial scale.
A marketing manager who uses AI to produce a first draft more quickly has improved productivity. A marketing manager who can interrogate a dataset, troubleshoot a website or build a simple software tool has expanded the boundary of the job itself.
The distinction becomes greater for individuals and small businesses. Organisations have traditionally dealt with complexity by employing specialists and creating departments. A sole trader, student, engineer, journalist or small-business owner has finite money, knowledge and time, and AI can reduce some of those constraints.
Productivity and Performance
Evidence that AI improves productivity is real but far from uniform. One of the best-known field studies followed 5,179 customer-support agents after the introduction of a generative AI assistant. Productivity, measured by issues resolved per hour, increased by 14% on average. The effect was much larger for novice and lower-skilled workers, who improved by 34%, while experienced and highly skilled workers saw smaller benefits. The researchers found evidence suggesting that the system was helping less experienced workers adopt some of the practices used by stronger performers.
A separate experiment involving 758 Boston Consulting Group consultants found substantial gains on tasks within GPT-4’s capabilities. People using the technology worked more than 25% faster, completed over 12% more tasks and achieved human-rated performance scores more than 40% higher. On a deliberately selected task outside the model’s capabilities, however, consultants using AI were 19 percentage points less likely to reach the correct answer than those working without it.
Both studies involved an earlier generation of generative AI than the systems available in 2026, so their precise productivity figures should not be treated as measurements of current models. Their broader findings remain useful: performance depends heavily on the task, the worker and whether the system is operating within an area where it can provide reliable assistance.
More recent evidence complicates the picture. METR studied experienced open-source developers working on repositories they knew well and found that allowing them to use early-2025 AI tools made them 19% slower. Before starting, the developers expected AI to make them 24% faster; after completing the work, they still believed it had made them around 20% faster. METR cautioned that the result applied to a particular group of experienced developers and a particular generation of tools, but the gap between perceived and measured productivity is difficult to ignore.
There is also a difference between saving time on individual tasks and producing measurable economic gains. Research into Danish workers across 11 occupations found widespread chatbot adoption and reported time savings, but no detectable effect on earnings or recorded hours within two years. The researchers estimated average time savings at about 2.8% of working hours among users, yet those gains had not translated into measurable changes in pay or working time.
AI can improve productivity dramatically in some circumstances, hinder it in others and save time without necessarily changing economic output. Effective use requires enough understanding to frame the problem, judge the answer, recognise when the machine has wandered outside its competence and decide what should remain a human responsibility.
The AI Leverage Gap
More than one billion people now use ChatGPT each week, according to OpenAI, but raw adoption says little about how much value any individual extracts from it. Two people can both say they use artificial intelligence every day while describing almost entirely different relationships with the technology.
One might use it to polish emails, summarise documents and answer occasional questions. Those applications can save time. Another might use the same underlying technology to research unfamiliar subjects, analyse documents, write software, interrogate data, test arguments, learn new skills, prepare commercial material and explore areas that previously sat outside their professional competence.
The first person has acquired a useful productivity tool. The second has expanded the range of work they can realistically attempt.
OpenAI’s enterprise data suggests that some users and organisations are moving towards deeper integration. Weekly messages across its Enterprise products increased roughly eightfold over a year, while use of structured workflows such as Projects and Custom GPTs increased 19-fold. As vendor data, those figures primarily describe activity within OpenAI’s own products rather than the wider economy, but they suggest a progression from occasional questions towards repeatable AI-assisted workflows.
The economically significant divide may therefore become a leverage gap. Having access to AI is increasingly unremarkable. The difference lies in how effectively people can turn that access into additional capability while retaining the judgement required to use it well.
Experience and Judgement
If AI makes specialist knowledge easier to access, experience might initially appear to become less valuable. In many occupations, its value may instead migrate towards judgement.
A machine can generate a contract, engineering explanation, market analysis, computer program or financial model. Someone still has to recognise whether it is any good, and that judgement comes from knowledge. Experienced people know where mistakes tend to hide. An engineer understands when an apparently sensible calculation violates physical reality. An experienced plant operator can recognise advice that would be dangerous on an actual machine. A journalist notices when a confident narrative is built on a weak source. A programmer understands why code that runs is not necessarily code that should be deployed.
AI can compress some of the work required to reach an answer without providing the experience needed to know whether that answer should be trusted. Verification becomes increasingly important as the volume and speed of AI-generated work increases.
The labour market may already be attaching greater value to those capabilities. PwC’s 2026 analysis of more than a billion job advertisements found that skills required in highly AI-exposed jobs were changing more than twice as quickly as in less exposed occupations. AI-exposed junior positions were also seven times more likely to request traditionally senior skills such as judgement and leadership.
That creates a difficult question for employers. Greater automation of routine junior work can improve immediate productivity while also reducing some of the work through which inexperienced employees traditionally developed expertise. The issue returns in education, where the same tension is already becoming visible.
Over-reliance and Critical Thinking
The easiest way to use AI is also potentially one of the least useful: ask it to think, accept what it produces and move on. Research involving 319 knowledge workers and 936 examples of real-world AI use found that greater confidence in generative AI was associated with less critical-thinking effort. The researchers found that AI changed the nature of cognitive work, moving effort away from producing material and towards verification, integration and oversight. That transfer can be valuable, but only when the verification actually happens.
The METR study provides a practical example of the same problem: experienced programmers believed AI had accelerated their work even when measurements showed they had taken longer. Confidence in an AI-assisted process and evidence that it is working are not necessarily the same thing.
A student who asks AI to explain calculus and then challenges the explanation has acquired an extraordinarily patient tutor. A student who asks it to complete the assignment has acquired an extraordinarily efficient way of avoiding learning calculus. The same principle applies in professional life.
Someone using AI to explore unfamiliar territory while checking sources, challenging assumptions and applying their own experience may steadily increase what they can do. Someone who routinely delegates thinking they previously performed themselves risks losing precisely the knowledge needed to recognise when AI is wrong. The technology can amplify intellectual curiosity or intellectual laziness, depending largely on the behaviour of the person using it.
Technology and the Value of Skills
Technological advances have rarely arrived without anxiety about what they would do to human skills and employment. Printing changed the economics of reproducing knowledge, mechanisation displaced established crafts, calculators raised concerns about the loss of arithmetic skills, and desktop publishing transferred work once performed by specialist typesetters and production departments onto an ordinary computer.
Those fears were not entirely misplaced. Technologies really did eliminate tasks, businesses and occupations, but they generally did not eliminate the need for people to perform useful work. Economic value moved elsewhere. Accountants no longer needed to spend hours performing calculations by hand, engineers no longer required a drawing office to produce every revision, and small companies could suddenly create material that once required professional typesetting and printing infrastructure.
Artificial intelligence belongs to that lineage, although its breadth sets it apart from many previous workplace technologies. A calculator automated calculation. CAD transformed technical drawing. Desktop publishing democratised publication. Search engines transformed access to information. Each concentrated primarily on a particular area of human activity.
AI can work across many of them during the same session. It can write, calculate, translate, analyse, research, explain, code and create visual material, while newer agentic systems can increasingly use software and undertake multi-stage tasks.
Previous technologies repeatedly made particular capabilities abundant and forced people, professions and businesses to reorganise around them. AI potentially applies that process across a much wider range of cognitive work at the same time. Learning to work effectively with intelligent systems, divide problems into suitable pieces, provide useful context, interrogate results and recognise when the machine should be ignored may prove more durable than expertise in any particular AI product.
Education and Early-Career Learning
Education and professional development depend partly on people doing difficult things before they become good at them. Essays, calculations, reports and examinations have traditionally served two purposes: they produce an answer while forcing the student to develop the knowledge required to reach it. Generative AI can now produce many of those artefacts without the learner acquiring the underlying understanding.
Used differently, the same technology can explain a concept repeatedly without impatience, change the level of an explanation, create examples, challenge an argument, generate practice questions and provide immediate feedback. A student can ask questions they might be embarrassed to ask in a classroom and continue until the subject makes sense. Individual tuition of that kind would once have been expensive or simply unavailable.
The problem extends into employment. Junior engineers, lawyers, programmers, journalists and managers traditionally learned partly by performing routine work, making mistakes and having those mistakes corrected by experienced colleagues. If AI absorbs much of that work, organisations may gain immediate efficiency while removing part of the process through which expertise is created. PwC’s finding that highly AI-exposed junior jobs are already more likely to demand traditionally senior capabilities suggests employers may have to reconsider how those capabilities are developed in the first place.
Education and employers therefore face a similar challenge: using AI to accelerate learning without allowing production of the answer to replace development of the understanding behind it. Teaching people to operate an AI product will not be enough if they lack the knowledge required to question what it produces.
Adapting to AI
Fear of artificial intelligence is understandable because the disruption is real. Jobs will change, some will disappear and skills that took years to acquire may lose economic value surprisingly quickly. Refusing to engage with the technology does not remove those pressures.
AI also allows individuals to move beyond some of the boundaries created by education, income, geography, employer, age or previous career choices. Someone who never learned programming can begin building useful software. A small business can access analytical capabilities once available mainly to larger organisations. An experienced worker can combine decades of practical knowledge with tools capable of researching, calculating, translating and producing at extraordinary speed.
The evidence does not support the idea that AI automatically makes people faster, smarter or more productive. Sometimes it does the opposite. Its value depends on the task, the technology and the person using it, while experience, curiosity and judgement remain essential to recognising when assistance has become dependence.
Yet AI also gives individuals access to knowledge and capabilities that would have been difficult, expensive or impossible for them to reach only a few years ago. The disruption is real, but so is the opportunity. People can be overwhelmed by that change, or they can learn how to use it to expand what they are capable of doing.
Key Industry Questions
- Will AI eliminate jobs?ย Some tasks and roles are likely to contract, while others will change or emerge. Stanford’s latest research found no evidence of widespread economy-wide displacement, although employment among 22 to 25-year-olds in highly AI-exposed occupations stood 19% below where it would have been had it kept pace with less-exposed peers by mid-2026.
- Does AI actually make workers more productive?ย Sometimes. Studies have measured substantial gains in customer support and suitable knowledge-work tasks, while METR found experienced developers became slower in a particular coding environment. Productivity depends heavily on the worker, task and AI system being used.
- Do AI time savings translate directly into economic gains?ย Not necessarily. Danish research found reported time savings among chatbot users without detecting corresponding changes in earnings or recorded working hours during the study period.
- Who can benefit most from AI assistance?ย The answer varies by task. In the customer-support study, novice and lower-skilled workers gained considerably more than experienced workers, suggesting AI can sometimes accelerate access to practices already understood by stronger performers.
- Can AI allow people to work outside their traditional profession?ย OpenAI’s usage data suggests people are attempting this at significant scale. It found that 43.5% of occupation-specific work messages, after generic tasks were excluded, concerned activities associated with another occupation. The data measures usage rather than the quality or success of the resulting work.
- Does using AI reduce critical thinking?ย It can when users place excessive confidence in the system. Research also suggests AI can shift cognitive effort towards verification and oversight, so the outcome depends partly on how the technology is used.
- Will professional expertise still matter?ย AI increases access to specialist capabilities while increasing the importance of assessing what it produces. Labour-market evidence suggests judgement and leadership are becoming more prominent requirements in some AI-exposed roles.
- What is the AI leverage gap?ย It describes the difference between having access to AI and using it effectively to expand personal capability. It is an editorial interpretation of emerging usage patterns rather than an established economic measure.
Strategic Takeaways
- AI adoption alone reveals little about the economic value users obtain from the technology.
- Productivity gains vary considerably by task, worker and AI system, and perceived gains do not always match measured performance.
- Cross-occupational AI use suggests individuals are attempting work that previously required greater specialist support.
- Domain knowledge remains important because rapid production increases the amount of material requiring judgement and verification.
- Employers and educators may need new ways to develop expertise if AI absorbs routine work traditionally used for training.
- The greatest individual gains may come from using AI to extend existing knowledge rather than replacing the effort required to acquire it.
