New Jobs AI Is Creating (With Evidence)

340,000 Americans worked as telephone switchboard operators at the job's peak. Then direct dialing arrived, and the job vanished. Not slowly. Gone.

Here's what didn't vanish: all the work that phone networks made possible. Telemarketing. Customer support. Telecommunications engineering. And eventually, everything built on voice communication that became the internet's backbone. The new jobs created by AI follow the same pattern, and the data backs it up.

The World Economic Forum's Future of Jobs Report puts a number on it: 85 million jobs displaced by AI and automation by 2025, but 97 million new roles created in the same window. Net positive: 12 million jobs. Nobody's leading with that headline.

This isn't cheerleading. It's pattern recognition. The same thing happened with the printing press, the automobile, and the internet. The technology always wins. The people who learn the new rules keep working. The ones who don't, don't.

The historical pattern nobody wants to admit

Every major technology in history has pulled off the same trick. It kills a job category, then quietly creates three more.

The printing press put professional scribes out of work. A scribe who copied manuscripts by hand wasn't going to survive when a press could do it in an afternoon. But the press also created publishers, editors, journalists, proofreaders, typesetters, and eventually librarians. It didn't shrink the world of written work. It exploded it.

The automobile killed horse-related trades: blacksmiths, farriers, stable hands, carriage makers. In exchange, it created mechanics, gas station workers, highway engineers, traffic police, auto insurance adjusters, and eventually the entire suburban economy that reshaped where and how people lived and worked.

The internet killed travel agents, video rental clerks, and print classifieds. It created web developers, SEO specialists, social media managers, UX designers, e-commerce logistics coordinators, and cloud infrastructure engineers. Most of those job titles didn't exist in 1995. By 2010 they were normal careers with career ladders and annual reviews.

Mobile added another layer. The entire app economy. Gig work. Influencer marketing. The job "app developer" barely existed in 2006. By 2015 it was a lucrative career path with its own bootcamps.

The pattern is not subtle. Destruction is visible and immediate. Creation is invisible until it isn't.

New jobs AI is already creating right now

Some of these roles have names. Some don't yet.

Prompt engineer was a joke title two years ago. Now it's a real position at serious companies, with salaries between $100,000 and $300,000 at the top end, according to data from job boards like LinkedIn and Indeed. Whether that specific title survives is debatable. That the underlying skill persists is not.

AI trainers are the humans who review AI outputs, flag errors, and teach the model what good looks like. It's tedious, granular work that requires genuine subject-matter expertise. A doctor reviewing medical AI outputs. A lawyer checking contract summaries. A writer grading creative quality. These jobs are being created at scale at companies like Scale AI and Appen.

AI ethicists and policy specialists are now employed at every major tech company and in government agencies. The EU's AI Act alone created demand for compliance roles that didn't exist 18 months ago. That's a category, not a niche.

Then there are the roles that don't have clean titles yet. The person who manages a company's relationship with its AI vendors. The specialist who audits AI outputs for a law firm before anything goes to a client. The human who handles the customer escalations that the AI chatbot couldn't resolve. These are real jobs happening right now, at real companies, carried out by people who didn't know that was their job description when they finished school.

Why "new jobs created by AI" always sounds unbelievable

Here's the problem with optimism about technology and employment. The jobs that get destroyed are concrete and visible. You can name them, count them, interview the people who held them. The jobs that get created are abstract and invisible until they already exist.

In 1994, if you told a laid-off travel agent that millions of jobs would be created around something called "search engine optimization," they'd have stared at you. If you told a video store clerk that entire studios would be built to create content exclusively for an internet streaming service, they'd have needed the term "streaming service" explained first.

We're in that exact same moment right now with AI. The new roles are forming in real time, and most of them don't have consensus titles yet. That makes them easy to dismiss. It also makes them easy to miss entirely.

Dmitry Kargaev addresses this directly in Don't Replace Me, calling it Rule #21: New Tech = New Jobs. Always. The historical record is consistent enough that you can treat it as close to a rule as economics offers. Not because technology is kind, but because human demand for things doesn't disappear when efficiency improves. It expands.

What the WEF numbers actually mean for your career

The 85 million displaced, 97 million created figure from the WEF isn't uniformly distributed, and pretending it is would be dishonest. Some industries face more displacement than others. If you want the realistic breakdown by sector, the ai job replacement statistics page goes through that in more detail.

The more useful question is: where do the 97 million new roles cluster?

According to the WEF, the growth is concentrated in: care economy roles (human-facing work that's hard to automate), green economy jobs (AI is being deployed in climate tech at scale), data and technology roles (obviously), and what the report calls "human-machine interface" roles: the jobs that exist precisely because AI exists and needs managing, auditing, directing, and interpreting.

That last category is the interesting one, because it's the least predictable and the fastest growing. Every AI deployment at scale creates overhead. Legal overhead, compliance overhead, quality-control overhead, customer-trust overhead. Humans fill that overhead.

The people best positioned for those roles aren't necessarily technical. They're domain experts who understand their field well enough to supervise AI in it. An accountant who understands what the AI got wrong. A nurse who can tell when the diagnostic model made an error. A lawyer who can spot when the contract summary missed a clause.

If you're worried about where you fit into this, the how to future-proof your career against AI breakdown is the practical version of this.

The "free from fruit duty" argument

There's a version of this that's less about job titles and more about what work actually feels like.

Imagine you spend 40% of your working week on tasks you find tedious. Formatting reports. Pulling data for presentations. Drafting first versions of documents you'll rewrite anyway. Scheduling and rescheduling. Researching things you already half-know.

AI does a lot of that. Not perfectly, not always, but well enough to save you hours a week.

That time has to go somewhere. Historically, it goes toward the work that actually required you in the first place. The judgment calls. The relationship maintenance. The creative problem-solving that nobody can automate because it depends on context that lives in your head and your history with the organization.

This is what Kargaev calls being "freed from fruit duty" in the book: the same way agricultural machinery freed humans from manually picking every individual piece of fruit, AI is freeing office workers from the manual equivalent. The work doesn't disappear. It gets redistributed toward humans doing things only humans can do.

Whether that's utopian or annoying depends on your tolerance for the tedious parts of your job. But the mechanism is real. McKinsey's research on automation consistently shows that even in roles facing significant automation, the displaced tasks represent a fraction of the overall job, not the whole thing.

What kinds of roles will exist in 10 years

Naming specific jobs is a fool's game, and anyone selling you a list of "the 20 AI jobs of the future" is guessing. But the categories are readable.

AI supervision roles will exist in every industry. Someone has to be responsible when AI outputs go wrong, and "the AI did it" is already not a legal defense in most jurisdictions. That liability creates human roles.

Human-facing work will grow relative to automated work. When every company's customer-facing AI sounds identical, the companies that offer a real human as a differentiator win a customer segment. That's not idealism, that's a market opportunity.

Domain experts with AI fluency will be the most valuable category in most professional fields. Not the person who understands AI best. The person who understands their field best AND can work with AI tools without friction. The gap between those people and domain experts who refuse to adapt will get wide fast.

Creative and taste roles won't disappear. AI generates. It doesn't discriminate well. Knowing what's good, what fits the audience, what will land, what to throw out: that's still human territory. If you want a longer argument for why, the breakdown of jobs AI can't replace makes the case.

The roles that don't have names yet are the ones that emerge when two things collide: a new capability and a real-world problem nobody fully anticipated. Every technology produces these. AI is producing them faster than any previous technology. Fast enough that some of them are showing up on job boards before the industry has agreed on what to call them.

The jobs AI is creating don't look like anything that existed before. Neither did "webmaster" in 1994, and that turned into a $100 billion industry.

Frequently asked questions

How many new jobs is AI creating?

The World Economic Forum's Future of Jobs Report estimates AI and automation will create 97 million new roles by 2025, while displacing 85 million, for a net gain of 12 million jobs. The distribution isn't even across all industries, but the aggregate number is positive. You can see the full breakdown in the AI job replacement statistics breakdown.

What new jobs is AI already creating right now?

Current AI-created roles include prompt engineers, AI trainers (human reviewers who grade AI outputs), AI compliance and ethics specialists, and AI-vendor relationship managers. Less formally, most professional roles are expanding to include AI supervision tasks that didn't exist three years ago. These often don't have clean job titles yet.

Will AI create more jobs than it destroys?

Historical evidence and current WEF data both point to yes, but with a caveat: the timeline doesn't line up neatly. Destruction happens faster than creation, and the new jobs aren't in the same place as the old ones. The net number is positive, but the transition involves real disruption for real people. If you're concerned about your specific situation, the will AI replace my job article gives a more personal framing.

What jobs will AI create that don't exist yet?

Nobody knows the full list, and anyone who claims otherwise is guessing. The identifiable categories are: AI supervision and audit roles, human-machine interface specialists, domain experts who manage AI in their field, and expanded human-facing roles in industries where automated alternatives feel impersonal. The switchboard operator's job disappeared, but the telephone created more jobs than anyone anticipated.

Do I need to be technical to get one of these new AI jobs?

No. The most in-demand category is domain experts who understand their field and can supervise or audit AI in it. That's an accountant, a nurse, a lawyer, a teacher. Technical AI skills help but aren't the entry requirement for most of these roles. The AI skills you actually need article covers what "AI fluency" looks like without a computer science degree.

Is optimism about AI and jobs naive?

It's evidence-based, which is different from naive. Every major technology shift in recorded history has produced more jobs than it eliminated. That doesn't make the transition painless or the disruption evenly distributed. But dismissing the creation side of the equation because the destruction is more visible is just bad pattern recognition. The printing press didn't shrink the world of written work. Neither will this.