The Future of Work: Will AI Take Jobs or Just Change Them?

It’s 8:57 on a Tuesday. Your inbox is already full, your calendar is double-booked, and someone’s asked for “a quick summary” of last week’s numbers. Then you notice something odd. The email draft is already written. The call notes have appeared on their own. The spreadsheet you feared is half-built, formulas and all.

This is how AI at work often arrives, not with a bang, but with quiet shortcuts that shave minutes off the day.

The fear is understandable. Recent reporting suggests around 65% of large UK companies plan to reduce staff by the end of 2026, and many are also bringing in AI-driven “virtual workers” to handle routine tasks. At the same time, a big chunk of roles won’t vanish, they’ll be re-shaped, with tasks shifting from doing the work to checking, directing, and improving it.

What AI really does at work, it doesn’t steal a job, it steals tasks

Most jobs aren’t one thing. They’re a bundle of tasks taped together: answering customers, updating records, drafting documents, chasing approvals, making judgement calls, spotting errors, calming people down. When people say “AI will take jobs”, what usually happens first is narrower. AI takes the easiest slices of the task bundle, the repeatable bits that follow patterns.

A few terms you’ll hear a lot:

• Automation: software that follows rules to complete steps (think “if X, then do Y”).
• AI assistants: tools that help you write, summarise, search, translate, and brainstorm. You drive, they support.
• Co-pilot tools: assistants built into the tools you already use (email, documents, spreadsheets, code editors).
• AI agents: systems that can plan steps and take actions across apps, often with human approval.

A helpful way to picture it is a busy kitchen. AI isn’t the head chef. It’s the commis who can chop, measure, and prep at speed. That changes who does what. The chef spends less time peeling potatoes and more time tasting, timing, and keeping standards high.

In practice, this often means:

• Less time spent drafting “first versions”.
• More time spent reviewing, adding context, and making decisions.
• Faster admin work, but also higher expectations for output.

The future of work isn’t only about replacement. It’s about task migration, where routine steps move to machines and people shift towards judgement, relationships, and responsibility.

Three kinds of work AI handles best: repeatable, predictable, and text-heavy

AI performs best when the work looks similar each time, the goal is clear, and the input is mostly text or structured data. That includes a surprising amount of “normal office life”.

Common examples:

• Sorting and summarising emails, then drafting replies in your tone.
• Turning meeting audio into notes, actions, and follow-ups.
• Producing weekly reports from standard templates.
• Booking shifts, managing rotas, sending reminders.
• First-line customer support for common questions.
• Basic translation of routine phrases and documents.

One widely shared estimate is that many roles could use AI to complete around a quarter of their tasks, especially the parts that are repetitive and text-heavy. Another often-cited view is that roughly 60% of roles may change tasks rather than become fully automated. That distinction matters. A job can stay “the same” on paper while feeling totally different day to day.

If your work includes lots of copying, pasting, checking, re-formatting, or writing the same thing in five different ways, AI will sit close to your desk.

Why AI agents change the game compared to chatbots

Chatbots talk. Agents do.

An AI agent is closer to a helpful junior colleague who can take a brief, complete steps across systems, and come back with a finished output. Often it operates with guardrails, such as approvals before sending emails, paying invoices, or updating a customer record.

The new workflow looks like this:

1. Human sets the goal: “Chase these invoices, then flag anything overdue by 30 days.”
2. AI executes steps: checks the finance system, drafts emails, schedules reminders, updates the CRM.
3. Human checks and signs off: approves messages, handles tricky cases, owns the outcome.

In everyday terms, that means fewer “tiny chores” clogging up the day. It also means mistakes can scale faster if nobody is watching. Agents can be brilliant at moving work along, but they still need human oversight, especially where tone, trust, money, or compliance is involved.

Which roles are most at risk in the next few years, and which ones get a boost

The safest way to think about risk is not by job title, but by task profile. Two people can share the same title and face very different exposure depending on how their job is set up.

A good rule of thumb:

• Higher risk: work that is routine, measurable, text-heavy, and easy to check.
• Lower risk: work that is physical, interpersonal, safety-critical, or full of messy edge cases.

The public conversation can swing between panic and denial, but the reality is more specific. Some roles will shrink quickly, especially where companies see direct cost savings. UK reporting also shows how local the impact can be. A BBC report highlighted how a million jobs in London could be changed by AI, with roles such as telemarketing, bookkeeping, and data entry heavily exposed (see BBC reporting on London jobs affected by AI).

Here’s a simple way to map it:

Role area	Why it’s exposed or boosted	Likely change
Admin and clerical	High volume of repeatable tasks	Fewer roles, broader responsibilities
Customer support	Lots of standard queries	Humans handle escalations and retention
Basic content	First drafts are easy for AI	Humans shift to editing and strategy
Entry-level tech	Routine coding and testing can be automated	Fewer “starter tasks”, higher bar
People-facing specialists	Trust and judgement matter	Demand holds or rises

Jobs with the biggest squeeze: admin, support, basic content, and entry-level tech

The tightest pressure often lands on roles built around coordination and routine production.

Admin and clerical work is exposed because so much of it is structured: scheduling, logging, updating, formatting, chasing, filing. In London, for example, job-change estimates have singled out bookkeepers and data entry specialists as vulnerable, because the tasks are pattern-based and easy to validate at scale.

Customer service is shifting as well. Chatbots and automated agents can handle password resets, delivery updates, appointment changes, and simple troubleshooting. The human work becomes the “hard bit”: complex complaints, safeguarding, sensitive situations, and keeping customers who are ready to leave.

Basic content is under squeeze too. AI can produce passable first drafts of FAQs, product descriptions, internal updates, and simple social posts. The value moves to voice, originality, fact-checking, and knowing what not to publish.

Entry-level tech is changing fast. AI can write boilerplate code, generate tests, and explain unfamiliar systems. That can reduce the number of small tasks that used to train juniors. It also pushes entry-level hires to show stronger fundamentals and better judgement earlier.

One stark point from current research is that around 6.1 million US workers sit in roles with high AI risk and low mobility, meaning they’re more likely to be displaced and less likely to have an easy path into a new field. The lesson for the UK is not that the same number applies here, but that the pattern can repeat when training and hiring pathways don’t keep up.

For a grounded look at how economists think about the trade-offs (productivity gains, displacement, and new work), see the Bank Underground analysis on generative AI and jobs.

Jobs that grow with AI: builders, fixers, trainers, and people-facing specialists

When AI becomes normal, work expands around it. Not everyone needs to become a machine learning engineer, but many organisations will need people who can build, control, and improve AI systems.

Roles likely to grow include:

• AI engineers and automation leads (building systems, connecting tools, managing agents)
• Data roles (quality, governance, labelling, pipeline health)
• Cybersecurity and fraud (AI raises the stakes on attacks and defences)
• Product and operations (designing workflows that mix people and machines)
• AI risk, compliance, and ethics (making sure tools are safe and lawful)

There’s also growth in work where trust and judgement are the product: healthcare support, care work, teaching support, skilled trades, complex B2B sales, and any role where the job is half technical and half human.

Many forecasts suggest AI could create more jobs than it displaces by 2030, but that doesn’t mean every community wins automatically. The winners will be places and firms that turn AI into better services and better roles, rather than just smaller payrolls. For an investment lens on productivity and knock-on effects, see Vanguard’s view on AI and the future of work.

How to stay employable when your job keeps changing

Employability in an AI-heavy workplace isn’t about knowing every tool. It’s about staying useful when the task mix shifts, and showing that usefulness in ways others can see.

Start with one honest question: if AI did your easiest 30% tomorrow, what would you do with the time?

That question is uncomfortable, but it’s also freeing. It points straight to where you can grow: the work that needs judgement, relationships, accountability, and taste.

Build your ‘human edge’: judgement, taste, relationships, and responsibility

AI can produce an answer. It can’t always tell you if the answer is wise.

Your “human edge” is the part of work that doesn’t fit neatly into a prompt:

• Judgement: choosing what matters, not just what’s possible.
• Taste: spotting what’s off, dull, risky, or wrong for the audience.
• Relationships: trust built over time, especially in tense moments.
• Responsibility: owning outcomes when the easy options fail.

Small examples show the difference:

• An AI-written report may look polished, but miss a key local factor (a strike, a supplier change, a new policy). You add the context and prevent a bad call.
• An AI customer reply may be “correct”, but cold. You rewrite it with empathy and keep the customer.
• An AI summary may hide uncertainty. You ask, “What evidence supports this?” and you check the source.

In other words, aim to become the person who can use AI without being fooled by it.

A useful reference point is how fast skills expectations are changing. Surveys and headlines show many workers are already planning to learn AI skills this year. Even if you’re sceptical of polls, the direction is clear (see TechRound’s report on learning AI skills in 2026).

A simple reskilling plan: learn one tool, redesign one workflow, prove one result

You don’t need a massive career reinvention. You need a repeatable loop.

1) Learn one tool (safely).
Pick the AI tool your workplace already allows (or one widely accepted for your role). Learn the basics: prompting, checking, citation habits, and what data you must never paste in. Keep a short list of “safe tasks” like summarising, drafting, and structuring.

2) Redesign one workflow end to end.
Choose a real process you do often, ideally weekly. Examples: preparing a client update, handling a case queue, producing a weekly report, onboarding a new starter. Map the steps, then decide where AI helps and where humans must stay in charge. Add review points like “human check before send” or “second-person sign-off”.

3) Prove one result in plain numbers.
Track something simple for two to four weeks:

• minutes saved per task
• fewer errors or rework
• faster response times
• higher customer satisfaction notes
• better consistency in outputs

Then write it up as a one-page “before and after”. That becomes a quiet portfolio you can use in reviews, interviews, or internal moves.

One warning that matters: follow your organisation’s rules on privacy, client data, and sensitive information. If you’re unsure, assume it’s not safe to paste in names, account details, health data, or anything regulated.

What companies and governments should do so AI doesn’t widen the gap

Individual effort helps, but the bigger outcome depends on how employers adopt AI and how public systems support transitions. If organisations treat AI as a pure cost-cutting tool, they risk turning good jobs into brittle ones.

The World Economic Forum has pushed one message repeatedly in its work on jobs: reskilling at scale is not optional. It’s the only way to stop productivity gains pooling at the top while everyone else competes for shrinking entry-level work.

If firms automate without training, they get short-term savings and long-term chaos

Cutting headcount can look neat on a spreadsheet, but the messy costs arrive later.

Quality can slip when nobody understands the process deeply. Customer trust drops when support becomes scripted and slow to escalate. Teams lose the “how we really do it” knowledge that keeps systems running. Then companies hire again, often at higher wages, to repair what broke.

A better approach is to redesign roles, not just remove them: move people from copy-and-paste work into quality checks, customer retention, exception handling, and process improvement. Pair AI use with clear review steps so errors don’t multiply quietly.

Reskilling that works looks like paid time, clear pathways, and real credentials

Reskilling fails when it’s a late-night hobby. It works when it’s treated as part of the job.

The basics are practical:

• Paid learning time, not just “optional training”.
• Clear pathways into AI-enabled roles (support to operations, admin to workflow specialist).
• Credible credentials that hiring managers trust.
• Targeted help for workers in highly exposed roles, so they aren’t last in line.

The goal isn’t to turn everyone into a developer. It’s to raise baseline AI literacy, create new internal roles, and keep human judgement close to high-stakes decisions.

Conclusion

AI will take some jobs, but it’s more likely to change more jobs than it deletes, at least in the near term. The sharpest cuts will hit roles built from routine tasks, and they may happen quickly where firms chase savings. For everyone else, the work will shift towards directing, checking, and owning outcomes.

The next step doesn’t have to be dramatic. Pick one task to automate, one skill to learn, and one result you can show in numbers. Do that, and you’re no longer waiting for the future of work. You’re shaping your place in it.

The office of tomorrow still needs human owners, people who decide what good looks like, and make sure the machines don’t forget it.