Your Job May Not Disappear. Half of It Might.

How AI Is Rewriting Work, One Task at a Time

A Next Horizon longread on which parts of work AI is taking over first, which careers are under the most pressure, and why the future of jobs is more complicated than “AI will replace everyone.”

AI is not replacing every job at once. It is quietly taking over individual tasks while humans remain responsible for judgment, priorities and outcomes.

AI Is Not Taking Jobs One at a Time

Picture a junior analyst starting work on Monday morning. A few years ago, the first hours of the day might have gone into pulling numbers from spreadsheets, cleaning data, searching through documents and building a first draft of a presentation. In 2026, an AI tool can do much of that before the analyst finishes a coffee. The analyst still has a job — but the job is already different.

That pattern is spreading. AI writes code, summarizes legal documents, generates advertising concepts, screens candidates, forecasts sales, translates meetings and handles routine customer requests. Yet offices have not suddenly emptied. The first big change is happening inside jobs: individual tasks are being automated, accelerated or handed to AI agents.

That is a more useful way to think about the future of work. A marketing manager does not simply “do marketing.” The job includes research, writing briefs, reviewing campaigns, negotiating with agencies, presenting to executives and deciding which risks are worth taking. AI may absorb several of those activities without becoming the marketing manager.

The best evidence so far points in this direction. The International Labour Organization estimates that roughly one in four workers worldwide is in an occupation with some exposure to generative AI, but says transformation is more likely than complete replacement for most jobs. Only a much smaller share of global employment falls into its highest-exposure category. ILO: Generative AI and Jobs — 2025 Update

A Job Title Hides a Bundle of Tasks

Job titles make work look cleaner than it really is. Most roles are a messy mix of routine work, specialist knowledge, judgment, communication and responsibility. AI does not need to master all of it to change the economics of the job. Taking over the repetitive 30 percent can be enough.

For a junior analyst, that might mean collecting data, cleaning spreadsheets and drafting the first summary. For a lawyer, searching case law and comparing clauses. For a software developer, boilerplate code and tests. For a recruiter, sourcing candidates and preparing outreach. The remaining work may be smaller in volume but higher in judgment.

That produces several possible outcomes. A company may need fewer people. The same team may simply produce more. Entry-level roles may become thinner while senior judgment becomes more valuable. Or cheaper production may create new demand and new kinds of work. “Automation” does not lead to one universal result.

A single job contains many different tasks. Routine digital work is increasingly automated, while negotiation, leadership, trust and judgment remain much harder to delegate.

Which Work Is Easiest for AI to Absorb?

AI has the clearest advantage when work is already digital, language-heavy, repetitive and easy to check. If both the input and the output live inside a computer, there is no physical barrier between the model and the task.

1. Clerical and administrative work

Data entry, scheduling, document formatting, standard correspondence, transcription, routine reporting and form processing sit near the front of the automation queue. The ILO still identifies clerical occupations as the most exposed group. That does not mean every administrator disappears; it means the repetitive layer of administrative work is becoming harder to justify as a full-time human workload.

2. Basic content production

Product descriptions, routine SEO copy, social-media variations, first-draft newsletters and generic visual assets can already be produced almost instantly. The scarce skill is shifting from making another piece of content to deciding what is worth saying, who it is for and whether the claim is actually good enough to publish.

That shift is already visible in advertising, where execution is becoming increasingly algorithmic and human value moves toward strategy, economics, positioning and judgment. See our updated Next Horizon analysis: AI Advertising: Automation, Generative Ads, Agentic Marketing and the Future of Paid Media

3. Entry-level analysis

A large share of junior knowledge work has historically existed because information was expensive to collect and slow to organize. AI is cutting that cost sharply. It can read folders of documents, compare scenarios and draft a first-pass briefing in minutes. Expertise still matters, but simply moving information from one format to another is becoming a weak career moat.

4. Customer support and routine sales operations

Routine support is another obvious target. AI can answer common questions, classify tickets, draft replies, qualify leads, summarize calls and update CRM records. Humans become more important when the situation stops being routine: an angry customer, an unusual exception, a negotiation, a sensitive complaint or any case where a confident wrong answer is expensive.

5. Parts of software development

Software development shows how quickly “assistant” can become “executor.” Coding agents can inspect repositories, write code, run tests and iterate on errors. But they are much better at well-specified, modular work than at deciding what product should exist, designing a resilient architecture or taking responsibility when a complex production system fails.

The Most Exposed Jobs Are Not Always the “Least Skilled”

Generative AI has produced an unusual reversal: some white-collar tasks are easier to automate than many physical ones. A language model can review a contract more easily than a robot can repair a leaking pipe. It can draft a financial memo more easily than it can safely move a patient, rewire a house or diagnose a strange noise inside an engine.

That is why exposure is not the same as disappearance. A role can be heavily affected by AI and still remain difficult to automate end to end. OECD research in 2026 makes the same distinction: many highly educated occupations are exposed because they contain lots of digital information work, but they also depend on judgment, social interaction and accountability.

The OECD also notes that most workers will not need to become AI engineers. Advanced AI-specific skills are likely to remain relevant to a small minority; for most people, the valuable combination is digital literacy, data interpretation, problem-solving, creativity, management and the ability to work effectively with AI. OECD: Skills in the AI Age (2026)

One of AI’s less obvious risks is the loss of junior work that once helped beginners gain experience and move toward senior roles.

Where the Pressure Is Building — and Where Demand May Grow

Nobody can produce a trustworthy list of exact job counts for 2035. Employment depends on wages, regulation, demographics, consumer demand and the broader economy, not only on technical capability. Still, the direction of pressure is already visible.

Roles and tasks under the most pressure

·         Data-entry and routine clerical roles, especially where work is structured and rules-based.

·         Basic bookkeeping and repetitive accounting support tasks.

·         Low-complexity copywriting, templated content production and commodity design work.

·         First-line customer support for standard questions and transactions.

·         Some junior research, reporting and document-review work.

·         Parts of translation and localization where context is limited and quality can be checked automatically.

Roles likely to grow or become more valuable

·         AI and machine-learning specialists, data engineers and infrastructure roles.

·         Cybersecurity and AI governance roles as organizations automate more critical systems.

·         Healthcare and care work, where demographic demand and human interaction remain powerful forces.

·         Education, training and mentoring roles that help people adapt to new tools and new work.

·         Skilled trades and field work that require dexterity, unpredictable physical environments and local judgment.

·         Roles that combine domain expertise with AI fluency: finance + AI, law + AI, marketing + AI, medicine + AI, engineering + AI.

The World Economic Forum’s Future of Jobs Report 2025 projected large churn rather than simple collapse: 170 million jobs created and 92 million displaced by 2030 across the macrotrends it studied, for a net gain of 78 million. Those numbers are survey-based projections, not destiny, but they underline an important point: technological disruption destroys some roles while creating demand elsewhere. World Economic Forum: Future of Jobs Report 2025

The Hidden Risk Is the Bottom Rung of the Career Ladder

The most uncomfortable problem may not be mass unemployment. It may be the disappearance of the work through which beginners used to become good.

Junior lawyers learn by reviewing documents. Analysts learn by building models and writing first drafts. Programmers learn by fixing small bugs and shipping simple features. Marketers learn by producing variations, reports and campaign assets. If AI absorbs too much of that work, companies can create a strange talent problem: they still want experienced people, but remove the training ground that used to produce them.

This is already visible in how workers themselves talk about AI. Anthropic’s 2026 Economic Index survey found that early-career respondents reported the highest perceived share of their work that AI could perform and were more concerned about job loss than more experienced workers. The survey is not representative of the whole labor market — it overrepresents AI users and technical occupations — but the pattern is worth watching.

The real management challenge is therefore not simply “automate as much as possible.” It is deciding which tasks should remain human because they build judgment, create accountability or train the next generation of experts. Anthropic Economic Index: Cadences (June 2026)

How AI Changes a Profession Without Erasing It

Marketing: execution gets cheaper, judgment gets more valuable

AI can already generate creative variants, summarize audience research, optimize bidding and interpret campaign data. That pushes the marketer’s value away from manual execution and toward harder questions: What should the brand promise? Which customers are actually worth acquiring? Is the growth incremental? Where should the algorithm be overruled?

Sales and revenue operations: prediction gets cheaper, decisions do not

AI can score leads, summarize calls and produce forecasts. But a forecast matters only when someone is willing to act on it — changing hiring, inventory, targets or spending. As prediction becomes easier to generate, deciding when to trust it becomes more valuable.

That distinction is central to our rewritten analysis of AI sales forecasting and revenue prediction, where the useful question is not whether AI can produce a number, but whether leaders know what to do with it.

Recruitment: search gets automated, accountability does not

Recruiters can use AI to search talent pools, summarize resumes, draft outreach and schedule interviews. The difficult part begins when a system starts ranking people. Hiring decisions affect livelihoods, can reproduce bias and increasingly face regulatory scrutiny. Automation makes transparent criteria and human accountability more important, not less.

We explore that problem in detail in AI in Hiring: How Artificial Intelligence Is Changing Recruitment.

The dividing line may not be between humans and machines, but between work that can be standardized and work that still depends on context, trust and responsibility.

What Becomes More Valuable as AI Gets Better?

The reflexive advice is often “learn to code.” Coding can be useful, but most workers will not build frontier models. They will use AI inside another profession. The durable advantage is knowing enough about the real work to direct the machine, challenge it and recognize when it is wrong.

1. Knowing what “good” looks like

AI can produce an answer quickly. Expertise tells you whether the answer deserves to survive. A senior accountant spots the assumption that breaks the model. A doctor notices a dangerous omission. A designer sees that the image is polished but wrong for the brand. Domain knowledge becomes the quality-control layer around AI.

2. Framing the problem

Prompting is not a secret language. The useful skill is defining the goal, supplying the right context, stating the constraints and deciding which parts of the problem should be delegated in the first place.

3. Checking the machine

As AI produces more first drafts, verification becomes a larger part of professional work: checking sources, reproducing calculations, testing assumptions and noticing when a plausible answer is simply wrong.

4. Judgment when there is no clean answer

Many important decisions are not optimization problems with one correct output. Should a company enter a market? Should a doctor order another test? Should a manager keep a brilliant employee who damages the team? AI can surface evidence and scenarios. Someone still has to own the choice.

5. Communication, trust and leadership

When routine information work becomes cheap, persuading, negotiating, teaching, reassuring, resolving conflict and taking responsibility in front of another person can become more valuable.

6. Learning faster than your job changes

The safest career may not be one particular title. It may be the ability to keep changing how you work. The people who benefit most from AI are likely to be those who absorb new tools quickly without handing their professional judgment over to them.

The Bigger Divide May Be Inside the Workforce

The future may not split neatly into “people with jobs” and “people replaced by AI.” A more plausible divide is between people who direct AI and people whose work is increasingly directed by software.

A skilled professional with capable agents may supervise work that once required a small team. That can make one person dramatically more productive — and can also concentrate leverage, income and decision-making in fewer hands.

Other workers may experience the opposite. In algorithmically managed roles, software can set schedules, measure output, score performance and narrow discretion. Both are forms of AI at work, but they feel very different to the person living inside them.

That is why the future of work is not only a technology story. The same system can remove drudgery, hollow out a role, expand autonomy or intensify surveillance. Much depends on how companies deploy it — and on who shares in the productivity gains.

What Changes First — and What May Take Much Longer

The near term: AI becomes ordinary inside knowledge work

The fastest changes will probably look boring. AI will disappear into email, spreadsheets, CRM systems, design tools, coding environments, document platforms and browsers. Productivity gains will come from fewer copy-paste steps, faster drafts, automatic follow-up and agents moving information between systems.

If agents become reliable: smaller teams, broader roles

If AI agents become dependable at multi-step work, one person may handle tasks that once required several specialists. Small companies could operate with much leaner teams. Job descriptions would broaden: less time performing every step manually, more time setting goals, supervising agents and handling exceptions.

The longer-term question: what does a “job” even mean?

The biggest change may eventually be organizational rather than occupational. Instead of hiring a person for every function, companies could assemble temporary combinations of people and AI agents around a goal. A freelancer could operate like a small agency. A doctor could oversee continuous monitoring for far more patients. A filmmaker could command tools that once required a post-production department.

That is the speculative edge of the story. Capability is only one constraint. Regulation, liability, trust, labor markets, energy costs and public acceptance can slow or redirect adoption. Long-range scenarios are useful for thinking, but they are not schedules.

As AI agents become more capable, one skilled professional may coordinate work that once required an entire team — shifting human value toward direction, review and decision-making.

The Better Question to Ask About Your Own Job

Some jobs will disappear. Some teams will shrink. Some careers that look stable today will become harder to enter. Pretending otherwise would be comforting, but not useful.

Still, asking only “Will AI replace my job?” hides the part that matters most. Jobs do not have an on/off switch. They are bundles of tasks, and those tasks are changing at different speeds.

A better exercise is to look at your own week. Which tasks are repetitive and digital? Which depend on context, trust or responsibility? Which teach you something valuable? Which would become more important if the routine work vanished?

The first wave of AI automation is not arriving as a humanoid robot walking into an office and taking someone’s chair. It is arriving as hundreds of small decisions: this report no longer needs to be written manually; this email no longer needs a first draft; this analysis no longer takes three hours; this support ticket never reaches a person; this junior task quietly disappears.

A profession can survive all of those changes and still become almost unrecognizable.

That is the real shift to watch. The value of producing the first acceptable answer is falling. The value of knowing what matters, recognizing when the machine is wrong, making a decision and taking responsibility for it is not.