AI Is Changing the Work. Is It Changing the Workers?
There is a familiar way of talking about artificial intelligence and employment. A new technology arrives, people worry that machines will take their jobs, companies promise that …
The interesting question is therefore no longer simply whether AI will replace workers. In many industries, we already have examples of AI replacing particular tasks and reducing the amount of human labor required to perform them. The more complicated question is what happens to the workers who remain. If a customer-service employee can handle twice as many conversations, does the company employ fewer people, or does it simply handle more customers? If a junior developer can produce code much faster, does that make the junior developer more valuable, or does it reduce the number of junior developers a company needs? If a marketer can produce ten pieces of content in the time it once took to produce one, does the marketing department become more productive, or does management simply expect ten times as much output?
Those questions matter because work is not just a collection of tasks. Jobs are also how people acquire experience, develop judgment, build careers and become senior enough to take on more responsibility. AI may therefore change something deeper than the work itself. It may change the path by which people become good at work.
The First Workers to Feel the Change Are Not Necessarily Being Replaced
One of the most useful real-world studies of generative AI came from researchers at Stanford and MIT who examined more than 5,000 customer-support agents at a Fortune 500 software company. The workers were given access to an AI assistant that suggested responses and provided information during customer conversations. The result was a roughly 14 percent increase in productivity, measured by the number of issues resolved per hour. The impact was considerably larger among less experienced workers, who saw productivity gains of up to around 35 percent.
That finding is more interesting than the usual "AI will make workers more productive" headline. The technology did not simply make the best employees better. It helped less experienced employees perform more like experienced employees by giving them access to suggestions based on the behavior and knowledge of stronger workers. The researchers also found improvements in customer sentiment and employee retention, suggesting that the technology could make a difficult job somewhat easier to perform.
There is an obvious positive interpretation. A new customer-service employee can learn faster, make fewer mistakes and become competent sooner. A company can provide better service without waiting months or years for every employee to accumulate experience. But there is another question hiding underneath the productivity improvement: if AI can help a new employee perform closer to the level of an experienced employee, what happens to the economic value of experience?
That question becomes particularly important when we move beyond productivity and start talking about headcount.
Klarna Shows What "Productivity" Can Mean to a Company
Klarna provides one of the clearest examples. The financial technology company introduced an AI assistant powered by OpenAI for customer support in 2024. According to Klarna's own filing, by the twelve months ending June 2025, the assistant was handling 69 percent of customer-service chats and doing work the company estimated to be equivalent to more than 700 full-time agents. Klarna also said the system contributed approximately $39 million in cost savings in 2024.
There is an important distinction here. Klarna did not simply discover that its employees could work faster. It changed the amount of human labor required for a particular category of work. The company itself described the AI system as doing the equivalent work of hundreds of full-time agents.
This is where corporate conversations about AI can become misleading. "AI makes employees more productive" sounds like a story about employees being given better tools. "AI allows the company to handle the same workload with fewer human workers" is a very different story. Both statements can be true at the same time.
For the customer, the difference may be invisible. They ask a question and receive an answer. For the business, however, the economics can change dramatically. If AI handles routine questions around the clock, a company may need fewer people performing repetitive support work. The remaining employees may deal with complex cases, escalations and customers who genuinely need human intervention.
That means the job does not necessarily disappear overnight. It can become a different job.
The Junior Worker Has More to Gain, and Potentially More to Lose
The productivity research around AI contains an uncomfortable paradox for younger workers. AI can make them dramatically more capable, but the same technology may also reduce the amount of entry-level work available to them.
Consider software development. Microsoft Research and academic collaborators conducted randomized field experiments involving 4,867 developers at Microsoft, Accenture and a Fortune 100 company. Across the combined experiments, developers with access to an AI coding assistant completed about 26 percent more tasks, with larger gains among less experienced developers.
On the surface, that sounds like excellent news for junior developers. Someone with limited experience can use AI to understand unfamiliar code, generate boilerplate, troubleshoot errors and move through problems faster. But imagine what happens if the economic logic continues. If a team previously needed five junior developers to perform a certain amount of work and AI allows three people to accomplish the same workload, the company may not necessarily hire five juniors again when those positions become vacant.
That creates a strange career problem. Junior employees traditionally learn by doing the simpler work first. They write basic code, prepare reports, conduct research, handle straightforward customer requests and gradually encounter more complicated problems. If AI absorbs a large portion of the simple work, people may have fewer opportunities to practice the fundamentals.
The technology could therefore make individual beginners more productive while simultaneously making the traditional beginner job less necessary.
That is one of the biggest workforce questions AI raises, and it is not solved by telling young people to "learn AI." They may absolutely need to learn it. But they also need opportunities to develop the underlying expertise that allows them to know when the AI is wrong.
Coding Is Already Becoming a Different Profession
Software development provides an unusually clear window into what this transition may look like because AI coding tools are already capable of performing substantial portions of programming work.
Anthropic analyzed 500,000 coding-related interactions involving Claude and Claude Code and found a significant difference between ordinary AI assistance and more autonomous coding. In its analysis, 79 percent of Claude Code conversations were classified as automation rather than augmentation, compared with 49 percent for ordinary Claude conversations. The research also found particularly high use in web development and user-facing applications.
That does not mean software developers are disappearing. It means the activity that defines the job is changing.
A developer who once spent much of the day writing code may increasingly spend more time describing what should be built, reviewing generated code, testing it, debugging failures and deciding whether the proposed solution is appropriate. The worker becomes less of a typist of code and more of a director, reviewer and problem solver.
Anthropic's own internal research offers a glimpse of this future. In a 2025 study of its engineers and researchers, employees reported using Claude for around 60 percent of their work and estimated an average productivity improvement of roughly 50 percent. At the same time, the researchers found that employees generally kept high-level thinking, design decisions and work requiring organizational context for themselves.
That last point may be one of the most important clues about the future of knowledge work. As AI becomes better at execution, humans may increasingly be valued for deciding what should be executed.
But AI Productivity Is Not Guaranteed
There is another reason to be cautious about simplistic predictions. AI does not automatically make every worker faster.
A 2025 randomized study involving experienced open-source software developers produced a surprisingly different result. The researchers expected AI tools to reduce task completion time, but the developers actually took longer on the tasks when AI was allowed, with the study reporting a 19 percent increase in completion time. The authors noted that the result may depend on the complexity and context of the work, but the finding is a useful reminder that AI assistance is not universally beneficial.
This matters because AI often looks more impressive in a demonstration than it does inside a messy organization.
A developer working on a clean example can ask an AI system to create a piece of code and receive an answer immediately. A developer working inside a huge, poorly documented legacy system has to understand whether the generated code fits the existing architecture, whether it creates security problems, whether it breaks something elsewhere and whether the AI misunderstood a dependency that only an experienced employee knows about.
The same applies to marketing, finance, law, HR and almost every other knowledge profession. The easier the work is to describe and verify, the easier it is to automate. The more the work depends on context, institutional knowledge, relationships and judgment, the more complicated automation becomes.
The Worker Is Becoming the Manager of AI
This may be one of the biggest changes to white-collar work. Instead of asking whether AI will replace the worker, we may eventually ask how many AI systems one worker will manage.
A marketing manager might use AI to generate campaign concepts, analyze customer feedback, produce first drafts, summarize research and create creative variations. A financial analyst might use AI to process documents, identify anomalies and produce an initial analysis. A lawyer might use it to review contracts and locate relevant clauses. A software engineer might delegate coding, testing and documentation.
In each case, the human remains involved, but the nature of involvement changes.
The worker becomes responsible for setting direction, checking quality, resolving ambiguity and making decisions. This sounds empowering, and often it will be. But it also raises the standard expected from the human. If an employee has access to powerful AI tools, management may reasonably ask why the employee is producing the same amount of work as before.
That is where productivity improvements can become workload increases.
If a writer can produce twice as much, the company may decide it needs twice as many articles. If a designer can generate ten concepts in an afternoon, the creative team may be expected to explore ten concepts rather than two. If a developer can build features faster, the product team may simply add more features to the roadmap.
Technology does not automatically create leisure. Historically, productivity gains have often created the capacity to do more.
The Worker's Identity May Change Along With the Job
There is also a psychological dimension that gets lost in discussions about productivity.
People often define themselves through what they are good at. A writer takes pride in writing. A designer takes pride in designing. A programmer takes pride in programming. A researcher takes pride in finding and understanding information.
When AI begins performing part of that activity, the question is not only whether the person still has a job. It is whether the person still recognizes themselves in the job.
A graphic designer who spends less time creating initial concepts and more time selecting and refining AI-generated options may still be a designer, but the craft has changed. A writer who spends less time drafting and more time researching, editing and developing ideas is still a writer, but the center of the profession has moved.
The Hollywood writers' strike offered an early example of this anxiety. During the 2023 dispute, the Writers Guild of America negotiated protections around the use of AI, including provisions concerning whether AI-generated material could be used as source material and whether writers could be required to use AI. The dispute demonstrated that the concern was not simply about losing jobs. It was also about preserving the role and status of the human writer within the production process.
The same tension appeared in the 2024 strike by U.S. video-game voice actors and motion-capture performers. Performers argued that AI could threaten their livelihoods through the use of synthetic replicas of their voices and performances.
These disputes are fundamentally about control over work. Who owns the output? Who gets paid? Who decides when AI can be used? And perhaps most importantly, where does the human contribution begin and end?
The Middle of the Organization May Change the Most
AI discussions often focus on entry-level employees because repetitive work is easier to automate. But the technology may also change the traditional middle of organizations.
Think about the people whose jobs involve collecting information, preparing first drafts, coordinating processes, producing reports and translating decisions from senior management into operational documents. These roles often sit between the person making the decision and the person executing it.
AI is increasingly capable of compressing that layer.
A senior manager can ask an AI system to summarize a large collection of documents. A salesperson can prepare proposals automatically. A consultant can generate an initial analysis. A marketer can turn a strategy into campaign variations. A manager can create a first draft of a performance review or presentation.
None of these capabilities necessarily eliminates the manager, consultant, salesperson or marketer. But it can reduce the number of people needed to support them.
That possibility is already influencing corporate restructuring. Reuters reported in 2026 that companies were announcing job cuts while shifting investment toward AI, with white-collar roles among those facing particular exposure. The same reporting noted that economists cautioned against attributing every layoff to AI because companies were also responding to broader economic pressures.
That caveat is important. Every layoff described as "AI-driven" is not necessarily caused by AI. Companies use AI as part of wider cost-cutting, restructuring and investment decisions. But the direction is still significant: executives are increasingly asking whether technology allows them to accomplish more with fewer people.
The Skills That Become Valuable May Not Be the Ones We Expect
For years, employees were encouraged to become specialists. Learn coding. Learn analytics. Learn design. Learn copywriting. Learn financial modelling. Learn another language. Become exceptionally good at a particular technical discipline.
Those skills will not suddenly become useless. In many cases, they will become even more valuable because someone needs to understand the work well enough to direct and evaluate AI.
But there may be a shift from simply possessing a skill to combining several skills.
The software developer who understands product strategy becomes more useful when AI handles more coding. The marketer who understands data, creative and business strategy can do more than someone who only knows how to write social posts. The financial analyst who understands the business behind the numbers is better positioned to evaluate an AI-generated report than someone who simply knows how to produce spreadsheets.
This is why the future worker may become more "full-stack" in the broadest sense. Anthropic's internal research found engineers using AI to work outside their traditional specialties, with backend engineers building interfaces and researchers creating visualizations.
AI does not necessarily make specialization disappear. It makes the boundaries between specialties easier to cross.
The Biggest Risk May Be Losing the Apprenticeship
Perhaps the most overlooked consequence of AI is what happens to learning.
People become experts by doing things badly before they learn to do them well. They write mediocre reports before they learn to write good ones. They create weak presentations before they understand what makes a strong one. They make bad marketing decisions before they develop judgment. They debug simple problems before they understand complicated systems.
If AI performs too much of the beginner work, the learning process changes.
A young marketer who asks AI to write every email may produce better emails today but learn less about writing. A junior developer who accepts every generated solution may deliver more code while understanding less about architecture. A new analyst who asks AI to interpret every dataset may complete assignments without developing the intuition required to recognize when an answer is nonsense.
This is why experienced workers may become more important, not less. Someone has to teach people how to question the machine.
The Real Divide May Be Between People Who Use AI and People Who Understand Work
There is a popular idea that the future belongs to people who know how to use AI. I think that is only partly true.
Knowing how to use AI will increasingly become a basic workplace skill, much like knowing how to search the internet, use spreadsheets or work with presentation software. It will be useful, but it will not necessarily be a lasting competitive advantage.
The more valuable skill may be knowing what to ask the AI to do, what not to ask it to do, how to evaluate the result and when the output is technically impressive but practically wrong.
That requires domain knowledge.
A person who knows nothing about marketing can ask AI to produce a marketing strategy. A marketer can look at the same strategy and notice that the customer assumptions are wrong, the positioning is generic, the measurement framework is weak and the proposed channels make no sense for the business.
The AI may have produced the better-looking document. The experienced person may still have produced the better decision.
The Workers Who Adapt May Not Be the Ones Who Work Harder
There is a temptation to frame adaptation as a personal productivity challenge. Learn the tools, use more AI, work faster and stay ahead.
I am not convinced that is enough.
The workers who benefit most may be the ones who understand which parts of their jobs should change and which parts should remain deeply human. They will use AI aggressively where it removes tedious work, but they will also protect the activities through which they develop judgment, relationships and expertise.
They will become comfortable delegating execution without delegating responsibility.
That distinction is important. If AI writes the first draft of a report, the human still owns the report. If AI generates the code, the developer still owns the system. If AI analyses the customer data, the manager still owns the decision.
The technology can change who does the work without changing who is accountable for the outcome.
AI Is Changing the Work. The More Interesting Question Is What We Become
The debate about AI and employment is often presented as a choice between two futures. In one, AI takes people's jobs and creates mass unemployment. In the other, AI makes everyone dramatically more productive and creates a new era of abundance.
Reality is likely to be considerably messier.
Some jobs will disappear. Some will become smaller. Some will become much more productive. Some will become more demanding because one employee will be expected to manage what used to require a team. Entire categories of work will probably be reorganized around AI rather than simply automated.
The most profound change, however, may happen inside the workers themselves.
A customer-service employee may become an exception handler rather than a person answering routine questions. A developer may become a system designer and reviewer rather than someone primarily writing code. A marketer may become an editor, strategist and director rather than someone producing every piece of content manually. A researcher may spend less time gathering information and more time deciding which information matters.
That could be a very positive future, but only if organizations take the development of people seriously. Productivity cannot be the only metric. If AI allows a junior employee to perform like a mid-level employee, we should also ask how that junior employee will eventually become genuinely mid-level. If AI handles the routine work, we need to make sure people still get opportunities to learn from doing it.
The most important workplace question of the AI era may therefore not be, "How many jobs will AI replace?"
It may be, "What kind of worker will this technology allow us to become?"
The answer will depend less on the technology than on the choices companies and workers make around it. We can use AI to eliminate the boring parts of work and give people more time for judgment, creativity, relationships and meaningful problems. We can also use it simply as a justification to reduce headcount, increase workloads and expect fewer people to produce more.
The technology does not decide which path we take.
People do.
