The Apprentice Cliff: Why AI Is Erasing the Next Generation of Experts
I've done a lot of talks and workshops on AI for marketing teams over the past year or so, and I've noticed something about the Q&A portion that I wanted to write about. A couple of years ago the questions were all about tools. What should we be using? Do you have prompt examples? What's company X doing that we're not? To the extent people were worried, they were sort of worried that a competitor was going to figure this stuff out first.
That's not what people ask me anymore. Now the questions are about jobs, and interestingly, they almost never come up during the session itself. They come afterward when nobody else is around. Is my role going to exist in five years? What am I supposed to tell the 24-year-olds on my team? A couple of people have asked me, and they were serious, whether they should tell their kids not to bother with marketing at all.
I'll tell you what I tell them, which is that I don't know. I wish I had something better than that.
I should say up front that AI has been great for me
Before I go any further, I should say that AI has been very, very good to me personally. I wrote a book about how the AI models decide which brands to recommend, and it made the Amazon bestseller list. I run workshops on it. I build software with it that I had no business building three years ago. I use it for research, for writing, for code, for sorting through data, pretty much all day long. There are projects on my plate right now that I would have said no to in 2022 because I didn't have the time or, frankly, the skills.
So I get it. I get why companies are moving as fast as they're moving. I'm one of the people benefiting from it.
What I've been paying more attention to lately is what all that speed is doing to the folks at the very start of their careers, and I really don't love what I'm seeing.
the numbers (and why I only half trust them)
Let me give you a few data points, with the caveat that you can poke holes in every single one of them.
In 2026, more than half of the layoffs that got recorded cited AI or automation as a primary reason. Stanford's "Risk Indicators" research found roughly a 13% decline in hiring for 22- to 25-year-olds in occupations with high AI exposure, and employment for older workers in those same fields held up much better.
Now, you can argue with any of that. The economy is part of it. Some companies are absolutely using "AI" as a nicer way to explain cuts they were going to make anyway. And I'm sure there will be a dozen more studies showing bigger effects, smaller effects, or no effect at all, depending on which jobs got measured and which months the researchers decided to look at. The number I care about is hiring, and the reason is pretty simple. A layoff leaves a record. Somebody had a job, they lost it, and there was an announcement or a filing or at the very least a LinkedIn post about it. A job that never gets posted in the first place doesn't leave a trace anywhere.
Think about it this way. Say your company normally hires seven junior analysts a year. Your existing analysts start using AI, they get through a lot more work, and so next year you hire four instead of seven. Nobody got laid off and nobody did anything wrong. If you asked the company about it they'd say, "Well, that's just workforce planning," and they'd be right. But three people didn't get their first job there. Maybe they got one somewhere else or maybe that somewhere else also hired fewer people that year. Nobody is ever going to write a story about it.
The Klarna thing
Klarna is my favorite example of how fast the AI story can flip, mostly because it was super public. In 2024 they announced that their AI customer service assistant was doing the work of roughly 700 people. You probably saw that number. I saw it in a dozen decks, in articles, in conversations about productivity for months. It gave execs a really clean thing to point at when they were talking about what AI meant for headcount.
By 2025 the CEO, Sebastian Siemiatkowski, was out talking about the problems. Quality had gone down, customers still wanted to talk to a human, and Klarna started hiring customer service people again.
That second story got nowhere near the attention of the first one, and I don't think it's because anybody hid it. "AI does the work of 700 people" is a headline. "Company rethinks the balance between automation and service quality and staffing levels" is a mouthful.
Marketing is running the same experiment
Marketing departments are doing their own version of this right now. A 2026 survey from Fractl found that AI now touches about half of what marketers do day to day. In that same survey, 48% of marketers said AI was making them less efficient, even though they were producing more. If you work in marketing you probably don't need a survey to tell you that. The amount of stuff we can crank out has gone through the roof. Five subject lines turns into twenty without any extra effort. One campaign concept turns into six. Research comes back in minutes, first drafts come back in minutes, and a deck that used to take days, multiple subject matter experts, and way too much coffee, can be put together in an afternoon.
It seems like expectations were reset almost instantly. Once twenty subject lines takes as long as five used to, guess what, people ask for twenty. Somebody still has to read all twenty. The six campaign concepts still have to get discussed and edited and argued about and narrowed down to one. The deck you built while eating your breakfast and browsing reddit at the same time still has to be checked and presented and then revised after the meeting. The time you saved doesn't sit there empty. Something else gets piled on top.
I use this stuff constantly, and I still have days where I got a ridiculous amount done, way more than I could have three years ago, and yet most of the day felt like sorting through output and fixing little things and deciding what deserved my attention. There’s no doubt I’m producing more, but I'm also less sure that I'm thinking better. My guess is companies will get better at this part. People figure out when to use the tool and when it's just generating more stuff than anyone wants, workflows settle down, and so on. That's a solvable problem, I think. The part I'm not sure is solvable is what happens to how new people learn. I've started calling it the apprentice cliff, and I think it's the bigger issue by far.
Where experts come from
Almost everybody who is really good at a job spent years doing that job badly first. That's just how it works.
A lot of my own early career was work I could do in an hour today. I took forever to write things. I researched topics I barely understood. I made decks that came back with more red ink than slide. I sat in meetings where I followed maybe 60% of what was going on and spent the rest of the afternoon trying to reconstruct the other 40% from my notes. At no point did I think of any of that as professional development. I was just trying to get the assignment done. It took me years to realize that a lot of the work I would happily have skipped was exactly where I learned the most.
Take a junior analyst who's been asked to build a report. Manually, it's three hours. With AI it's thirty seconds and the output is almost certainly 100% correct. That doesn't look like a hard call. But think about what happens in those three hours. The analyst has to sit with the data. Two numbers don't line up. They have to go figure out why, and in the process they find out that two departments define the same metric differently. So they go ask which one is right, and they get the annoying answer, "Well, it depends on what the report is supposed to measure." A few months later they're the first person in the room to spot that problem, and nothing about it shows up in the finished report. There was no training module for it. Nobody scheduled a lesson on conflicting metric definitions. They learned it because the assignment forced them to work through something they didn't understand.
There's a Yale marketing professor, K. Sudhir, who has written about the difference between effort and practice, and I think that's the useful distinction here, because from the outside the two look like twins (identical… not fraternal). There's plenty of pointless effort in every job. I absolutely don’t miss spending hours fixing PowerPoint formatting. Nobody needs to copy numbers from one spreadsheet into another for character-building purposes. Bad process is bad process, and AI is very good at getting rid of it.
The hard part is telling which tedious tasks are only tedious and which ones are giving somebody the muscle memory they need to earn. I'd argue most managers can't tell either, and when I say managers I'm including myself here, because we're looking at the output and not at the learning.
Same thing with a junior copywriter. They take a couple hours on a draft a senior writer would knock out in thirty minutes, and AI does an acceptable version in seconds. From the company's side the math is pretty dang clear. What that writer looks like in five years is a lot harder to measure, and nobody quite knows what got lost when those two hours went away. Maybe they use the time to study better work and test more ideas and get good faster. Or maybe they get very good at editing AI copy and never learn how to build an argument from a blank page. We're going to find out by living through it. You can run the same thought experiment on research, strategy, design, code, analysis, and a bunch of other fields I don't know well enough to have an opinion on.
Junior people are supposed to be inefficient
Junior employees are slow partly because they're learning. They stop to look things up. They ask questions that sound basic. They make decisions a more experienced person wouldn't, and in some cases those outside and fresh decisions solve problems in new ways. Somebody has to explain the work, correct it, and sometimes send it back. All of that costs time, and as far as I can tell it's also the only process anyone has ever come up with for producing experienced employees.
Boston Consulting Group put out a report recently saying more than half of senior execs think young talent is developing too slowly. I would love to know how many of those same executives have also redesigned entry-level roles around efficiency, cut junior headcount, or told their senior people to use AI instead of delegating. Can I prove that connection? No. Do I think it's there? One billion percent.
Companies have been trying to get rid of low-value work for decades, and AI lets them get rid of much more of it, much faster. My issue with the label is that "low value" is a statement about what the task does for the company today, and the thing it leaves out, because nobody measures it, is what the person doing the task picks up along the way.
The timeline that worries me goes something like this: A company decides not to hire a 24-year-old analyst in 2026. That year, there isn’t any visible downside at all. The work gets done, costs are lower, the remaining team looks more productive. In 2031, that business has one fewer person with five years of experience in it. One missed hire doesn't matter. A few thousand companies making the same call for the same reasons adds up to quite the mess.
Now, I could be wrong about all of this. I’ve certainly been wrong before. I’ll be the first to admit that. There are a few ways it could go fine, too. One, career paths change enough to absorb it. Two, AI turns out to help junior people learn faster than the old way did, which is possible, I just haven't seen it. Three, companies get serious about formal training once informal on-the-job learning stops working. Four, middle management shrinks along with the entry level, and companies just need fewer experienced people than they used to. I'm open to any of those.
What I can't get around is that companies are still going to need people who can make a judgment call when the right answer isn't obvious. Somebody who can look at a perfectly accurate analysis and say, "Yeah, but that recommendation is a bad idea." Somebody who knows what the client is asking for when the client hasn't explained it well, who can tell when a number smells wrong, who knows which problem can wait and which one blows up if nobody deals with it right away. I learned most of that by getting it wrong first. There might be a faster way to build that kind of judgment. I just haven't seen it.
Human-made is the new organic (maybe)
The other thing I've been thinking about is what an unlimited supply of AI-generated work does to the value of the human-made kind. A few early signals, for whatever they're worth. Coca-Cola and Valentino both got hammered for AI-heavy campaigns that people thought looked cheap. Erie Insurance ran a campaign around the line "No AI bodies or people. Just real people," and it became their most-liked Instagram post of the year. DoubleVerify says 43% of North American consumers think intrusive AI-generated ads hurt their perception of a brand. The Authors Guild is launching a certification for human-written books.
Some of that is probably a phase. Every new technology goes through a period where people are fascinated and annoyed at the same time, and AI creative is going to keep getting better until most of it is impossible to spot. At some point AI involvement is just going to be assumed in most commercial creative work. My hunch is that human involvement gets more valuable at exactly that point, once it stops being the default.
We've seen this play out before in a few different ways. Industrial food production created the premium market for organic and local. Streaming made music essentially unlimited and people went out and bought vinyl anyway. Digital cameras made a technically fine photo trivial, and suddenly film was interesting again to anyone who cared about the process. The mass-produced version won in every case, but what changed was what a certain slice of the market was willing to pay extra for. I can’t take credit it for it, or even properly source it as I’ve yet to find it’s origin, but I’ve been using the phrase "human-made is the new organic" as shorthand for this idea. If AI makes acceptable creative work cheap and infinite, some buyers are going to start caring about who made the thing, how they made it, and whether a human was involved in any way that mattered.
Which puts companies in kind of a weird spot. Right now they're being rewarded for pulling people out of creative and intellectual work, and in the short term the economics hold up. But the people getting pulled out are the same people who would have spent the next ten years accumulating the knowledge that makes an expert.
We already know how to make experts
Last thing I'll say on this. We're not missing a process here. We know how to produce experienced people. You hire them before they're ready, you give them real work, you correct it, and over time you trust them with more. It takes years and it doesn't always work out. A manager can spend an afternoon explaining something they could have done themselves before lunch and then lose that person a year later, right as they were getting useful.
AI makes it very, very easy for that manager to skip all of it and just do the work. Most managers will, when the deadline is tight and the team is already stretched thin. I would, and I have. Ten years of that decision, made over and over and over, changes who's around to run the department when you're gone.
Look at the people who might realistically be running your department five or ten years from now. What are they doing today that's getting them ready for that? What are they still allowed to get wrong? Who on your team has enough time to explain what went wrong when they do?
Jarred Smith is the author of Explainable: Why AI Recommends Some Brands & Ignores Others, an Amazon bestseller on AEO, GEO, and SEO. He's a marketing leader with nearly 20 years of experience across healthcare, public media, retail, and environmental services. Find him at jarredsmith.com.