The new jobs are coming. Nobody has built the road to them yet.

Two reports about AI and work came out within a few days of each other last week, and the headlines made them sound like opposites.

The first was the monthly count of job cuts from Challenger, Gray & Christmas, an American outplacement firm. AI, it said, is now the reason US employers give most often for cutting jobs this year.

The second was a long study from the McKinsey Global Institute. Over the next ten years, it said, the US could gain more jobs than automation takes away.

So which is it? Are the machines coming for the work, or is there more work on the way than ever?

Read past the headlines and the two reports argue less than you’d think. Neither says the work is running out. What the second one says, over 82 pages, is that the hard part is getting people from the jobs that shrink to the jobs that grow. The new jobs may well be there. Most of the roads to them aren’t.

Fewer job cuts, and more of them put down to AI

Start with the count. From January to September this year, US employers announced 573,195 job cuts. That’s 39% fewer than in the same nine months of 2025, and September’s total was the lowest for any September since 2022.

But employers said 120,136 of those cuts were down to AI. That’s 21% of the total, more than any other reason. Over the same months last year, AI was given as the reason for 17,375.

Fewer job cuts. More of them put down to AI.

Job cuts announced by US employers, January to September

2025, every reason 946,000

2025, put down to AI 17,000

2026, every reason 573,000

2026, put down to AI 120,000

Two things are worth knowing before that blue bar worries you.

The first is that 2025 was a strange year. Nearly 294,000 of its cuts were put down to DOGE, the US government’s cost-cutting drive. Take those out and last year’s total looks much closer to this year’s.

The second is what Challenger actually counts. It counts announcements, and it records the reason the employer gives. A cut that’s announced isn’t always a cut that’s made, and the reason given isn’t always the whole reason. “We’re investing in AI” is a much easier thing to tell investors than “sales are slow”. It works the other way too: some cuts that AI really did cause will have been filed under something else.

So the blue bar measures what companies say about their cuts. That’s worth knowing. It just isn’t the same as measuring what AI is doing to work.

The bigger number is 11 million

McKinsey’s study, published on 29 September, asks a different question. Not how many people lost their jobs this year, but how the whole shape of American work might shift over the next ten.

Its central estimate goes like this. Automation could cut the work the US economy needs by about 21% of today’s working hours. That’s the equivalent of about 36 million jobs. At the same time, an ageing population, rising living standards, a building boom, the tech industry and AI itself could create demand for about 41 million.

More jobs than it takes away, then. The interesting part is what happens to the people in those 36 million jobs.

For about 25 million of them, McKinsey expects growth in their own line of work to cancel out the automation. They stay where they are, in a job that changes around them. The other 11 million, about 7% of everyone working in the US, may need to change occupation altogether. That’s the middle of a wide range. If automation spreads faster, it could be 16 million. If it spreads more slowly, about 6 million.

The report sums up the whole problem in one line: “The next decade’s challenge is mobility, not scarcity.”

Here’s why moving is the hard part. In a normal year, about 215,000 Americans switch from one broad group of jobs to another: out of retail and into healthcare, say, or from the office to a building site. To move 11 million people in ten years, about 770,000 would need to make that kind of switch every year. That’s 3.6 times the usual number.

Ten years at pandemic pace

US workers switching from one broad group of occupations to another, per year

A usual year 215,000

2019 to 2022 788,000

Needed every year to 2035 770,000

The US has kept up that sort of pace once before. During the pandemic, about 788,000 people a year changed occupational group, and the country got through it without the lasting damage many feared. But that was a shock that lasted two or three years. This would be ten.

Who has to move

The moving isn’t spread evenly. More than three quarters of the people McKinsey expects to change occupation work in just three of its 22 groups: office and admin support, retail and sales, and transport and logistics.

Most of the moving starts in three kinds of work

Where the 11 million people who may need a new occupation by 2035 work now

Office and admin support 4.8 million

Retail and sales 2.3 million

Transport and logistics 1.4 million

Food service 0.9 million

Everything else 1.7 million

Narrow it down further and about a third of them are in five jobs: customer service representatives, retail sales assistants, office assistants, cashiers and warehouse workers. If you do one of those, this report is talking about your job.

It also falls hardest on the people who earn least. McKinsey reckons lower-paid workers are 7.6 times as likely as higher-paid ones to need a new occupation. People without a degree, people under 25 and women are all more likely to be on the move too, because of the jobs they’re more likely to be doing.

And the growing jobs aren’t next door to the shrinking ones. The biggest growth is in healthcare support, the healthcare professions, construction and management. A cashier is good with customers, reliable, and quick on a till. The new demand is for caring, building and running things. The skills don’t line up neatly, and that’s the whole matching problem in a sentence.

Three kinds of road

This is the part of the study we found most useful, because it’s more honest than most. Rather than asking whether a worker could move into a growing job, it asks how hard the move would actually be.

McKinsey grades every route on four things: whether the job at the other end is growing, how many of your skills carry over, whether the pay holds up, and how long it takes to get any licence or qualification the new job needs. Then it sorts the routes into three kinds of road.

• A direct road has lots of skills in common, no pay cut, and less than six months of training.
• A winding road has a fair few skills in common, a small pay trade-off at worst, and up to two years of training.
• An unpaved road is everything else: big gaps in skills, a possible pay cut, and no clear end to the qualifications.

Only one in seven has a direct road

The 11 million moves, by how hard the road is

14%Direct 41%Winding 45%Unpaved

How many have a direct road, by what they earn

Lowest-paid fifth 10%

Highest-paid fifth almost 40%

Only one person in seven has a direct road. Nearly half are looking at an unpaved one.

Again, it depends on what you earn. Only 10% of people in the lowest-paid fifth, on less than $38,000 a year, have a direct road. In the highest-paid fifth it’s almost 40%. The people with the furthest to travel have the worst roads.

Qualifications are a big part of the reason. About 85% of the growing jobs need some kind of certificate or licence. And 84% of the growing occupations ask for education beyond school, against 45% of the shrinking ones. A licence can turn a short road into a long one, especially if you have to keep earning while you get it.

There’s one piece of good news hiding in here. Most of these routes don’t mean a pay cut. McKinsey reckons only about 3% of the people who move, roughly 305,000, would end up earning less. Most moves are up or level, as long as people can make them.

Three journeys, and what our scores say about them

McKinsey walks through three real routes in detail. We’ve set them beside the scores this site gives the same jobs. On manvbot, a job’s score is how much of the everyday work still needs a person, out of 100. It’s built from the International Labour Organization’s ratings of what today’s AI tools could do with each task.

The move	Road	Skills in common	Pay	Our score, from → to
Dishwasher → home health aide	Unpaved	20%	About 12% more	87 → 75
Packer → assembler	Winding	52%	About 10% more	80 → 68
Office assistant → project manager	Winding	58%	154% more	41 → 58

The first is the hardest. A dishwasher already knows about hygiene and lifting, which a home health aide needs too, but has to learn the caring itself: helping people move, checking their vital signs. Getting certified takes about a month of formal training, and employers often prefer the longer nursing assistant course. McKinsey thinks around 300,000 people could be on this road by 2035.

The packer has more in common with the job at the other end: lifting, quality checks, often a forklift. What’s missing is the fine, fiddly work of assembly. About 170,000 people could take that road. The office assistant already plans, schedules and keeps things moving, but may need a degree and a project management qualification. That’s 120,000 people, and the biggest pay rise of the three, after the biggest investment.

Now look at the last column, because it surprised us.

Kitchen helpers and hand packers score 87 and 80 here, which puts both in our top band: strongly holds up against AI. Today’s AI tools can’t scrub a pan or pack a box. Yet McKinsey has both jobs on its list of shrinking ones, and their roads lead to jobs that score lower on our measure, not higher.

That isn’t a contradiction. Our score measures AI software. It says nothing about dishwashing machines, packing lines or warehouse robots, and the job pages say so. McKinsey is modelling all of those, plus changes in what people buy. Its own report makes the same point plainly: “Exposure and transition risk are not the same.”

So a high score here isn’t a promise, and a low one isn’t a sentence. General office clerks, on 41, have one of the lowest scores on the site, and a road out with a big pay rise at the end of it. It’s why every job page here says the same thing next to the number: it’s not the odds you keep your job.

Most people won’t change jobs. Their jobs will change.

Eleven million is the number that made the headlines, but it’s the smaller story. Remember the 25 million whose jobs are expected to change around them. Add everyone else whose work shifts, and McKinsey thinks more than 70% of US workers will see at least 15% of their working time move to different tasks.

For 28% of workers, less than 15% of their time gets reshuffled. For 47%, between 15% and 30%. And for a quarter, it’s more than 30%: their day will look substantially different.

Cashiers are McKinsey’s example of a job that shrinks and changes at the same time. It expects US cashier jobs to fall from 2.9 million to 2.4 million by 2035, counted as full-time equivalents. But the ones who stay won’t be doing a smaller version of the old job. The scanning goes to the machines. What’s left is sorting out payments that go wrong, helping customers and keeping an eye on the self-checkouts: the parts a person still does better.

Truck drivers go the other way. McKinsey expects their numbers to rise, from 1.3 million to 1.5 million, even as AI takes on the paperwork, the routine messages and some of the route planning. Driving and handling the load become a bigger share of the day, not a smaller one. Self-driving trucks could change that one day, but McKinsey doesn’t expect them to outweigh the growth over the next ten years. Our truck driver page scores the job 76, and points out that self-driving trucks are a robot question it doesn’t measure.

The skills that travel

If the roads are the problem, skills are what gets you down them. McKinsey sorts the useful ones into three kinds.

Essential skills are the ones most jobs want: solving problems, leading, communicating, managing people and processes, and paying attention to detail. More than three quarters of occupations ask for them. They won’t make you rich, but they widen the number of doors you can walk through.

Enabling skills are the ones that lead to better pay: making decisions, coming up with new ideas, thinking critically. They’re concentrated in higher-paid work.

Empowering skills help you keep up as the work changes, and employers have started asking for them much more. In job ads, demand for AI fluency, meaning being able to work well with AI tools, went up about elevenfold between 2022 and 2026. Demand for adaptability went up about fivefold, and for willingness to learn, resilience and curiosity it roughly tripled.

The report adds a sensible warning about that first one. Today’s AI tools will be replaced by others, so AI fluency isn’t somewhere you arrive. The skills that help you learn the next set of tools matter just as much as knowing this one.

What you can do with this

If your job is on McKinsey’s list, the report isn’t a verdict. It’s a map with some of the roads missing. A few things help.

• Look at your tasks, not your title. The routes that work are the ones that take your skills with them. Every job page here breaks the work down into tasks, so find yours and see which parts of your day would travel.
• Look for the short roads first. A move that keeps most of your skills and needs a short course beats a leap that needs a degree. You can set two jobs side by side, customer service and aged care for instance, and see where the work overlaps.
• Count the qualifications early. Most growing jobs ask for one. Find out how long yours would take, who pays for it, and whether you can earn while you do it.
• Get good with the tools in the job you have. For most people the change will come to them, not the other way round. Being the person who’s good with the new system is a skill that travels.

And if you run a business, a college or a government department, the report’s last two chapters, on lowering the barriers and helping people make the move, are written for you. Most of these roads are missing a few planks, not the whole bridge. Somebody has to lay them.

The new jobs are coming, and on McKinsey’s numbers there’ll be about 16 million openings for the 11 million people who may need to move. There’s work at the end of the road. Whether people can get there is the part that’s still up to us.