Which jobs will AI change the most, and which skills protect you?

"Will AI take my job?" is the wrong size of question. A job is a bundle of tasks: answering emails, updating records, talking to customers, making judgment calls. New AI tools speed up some of those tasks a great deal, leave others untouched, and create a few new ones. The useful question is which of your tasks will change, and what you will do with the time.

Three recent sources give a grounded picture: a large employer survey from the World Economic Forum, an OpenAI and University of Pennsylvania study of task exposure, and one of the first field studies of a chat assistant used at work.

What employers expect

The World Economic Forum's Future of Jobs Report 2023, published at the end of April, surveyed 803 companies employing more than 11.3 million people in 45 economies. Their expectations for 2023 to 2027:

Fastest-declining roles are mostly clerical, such as data entry clerks, cashiers, bank tellers and postal workers. Employers foresee 26 million fewer record-keeping and administrative jobs.
Fastest-growing roles are led by AI and machine learning specialists. Sustainability, business intelligence and information security specialists come next.
AI specifically is expected to be adopted by nearly three in four companies. Half expect it to add jobs in their organization and a quarter expect it to cut jobs.
Overall, employers expect about 23% of jobs to change through creation and loss, with a small net decline of about 2%.

Keep the limits in mind. These are forecasts by employers, not measurements. Companies also overestimated automation last time: in 2020 they expected nearly half of business tasks to be automated within five years, and today they estimate machines handle about a third.

Which tasks are most exposed

The "GPTs are GPTs" study from March took a task-level view of 19,265 tasks across more than 1,000 occupations. It found that tasks involving programming and writing are the most exposed to language models, while tasks that depend on science and critical-thinking skills are less so. It also found that higher-wage jobs tend to be more exposed. The authors stress that exposure means a tool could do a task much faster, not that anyone will be replaced.

A field study published in April shows what this looks like in practice. Erik Brynjolfsson, Danielle Li and Lindsey Raymond followed 5,000 customer support agents as a company rolled out an AI chat assistant. Agents with access resolved 14% more issues per hour on average. The largest gains went to newer and less-skilled agents; the most experienced agents saw little change. Customers were also happier and fewer agents quit. That is one company and one kind of job, but it suggests AI often works as a fast way to pass on the know-how of the best workers.

Which skills protect you

The Forum's survey names analytical thinking and creative thinking as the most important skills for workers, and the two growing fastest. Technology literacy is the third-fastest riser, and among attitudes, curiosity and lifelong learning and resilience and flexibility are gaining the most. Training workers to use AI and big data is the third-highest training priority for the companies surveyed.

Read together with the task studies, a pattern emerges. The protective skills are the ones that sit around the AI: framing the problem, judging whether an answer is right, dealing with people, and adapting when the tools change.

A worked task map: an office administrator

Here is one job broken into tasks. The time shares and ratings are illustrative, to show the method, not data from the studies.

Task	Share of week	Likely AI role	Skill to grow
Drafting routine emails and notices	20%	Drafts; you edit	Clear briefs, editing
Entering invoices and updating records	25%	Largely automated by software	Checking data, spotting errors
Scheduling meetings and rooms	10%	Assists	Prioritizing
Answering staff questions about policies	15%	Drafts answers from documents	Knowing when the answer is wrong
Handling upset callers and visitors	10%	Little	Empathy, de-escalation
Organizing events and suppliers	15%	Brainstorms options	Negotiation, planning
Training new colleagues	5%	Helps write guides	Teaching

In this example, about 60% of the week is in tasks where AI can draft or automate, and about 40% depends on people skills and judgment. The person in this job is not about to vanish, but the mix is moving toward the right-hand column.

Try it: Copy the table and fill it in for your own job, using your last two weeks as the reference. Then circle the one task where AI could save you the most time and the one skill on the right you most want to build this year.

How do you map the tasks in your own job?

Start from evidence, not from your job description. Look at your calendar, sent emails and to-do lists for two typical weeks and list every recurring task in plain verbs: "write," "check," "call," "decide." Most people end up with 10 to 20 tasks.

Then estimate the share of your week each one takes and rate it on two questions. First, is it mostly about producing or sorting text and data? Those tasks are the most exposed. Second, what happens if it goes wrong? Tasks with high stakes, such as legal, medical or financial decisions, need a person to check them even when AI does the first draft.

Finally, add the right-hand column: for each exposed task, the skill that would let you do it better with AI, and for each protected task, the skill that makes it more valuable. Repeat the exercise every six months or so. The tools are changing quickly, and the map will too.