Teams across the market are smaller than they were two years ago and getting more done. Almost all of them have the same hole: nobody junior. That gap didn't start with AI. Long before generative AI, a lot of companies were already avoiding junior hires, because junior people are difficult and expensive to develop: years of mentoring, tolerance for mistakes, a willingness to pay someone to learn on the job. The youth unemployment numbers have been ugly for a while because of it; recent college graduates were already unemployed at roughly 5.6%, well above the 4.2% national rate, before you even get to AI. AI isn't creating that problem. It's making it worse, because the junior work, drafting the first pass, pulling the data, writing the routine ticket response, is exactly the work AI now does before a human touches it. The debate about AI and jobs keeps getting framed as one question: does AI take jobs, yes or no. That framing is already wrong, and the data coming out this year says so plainly. There are two different futures happening at once, for two different kinds of work, and the gap between them is the actual story.
What the Numbers Actually Show
PwC's 2026 Global AI Jobs Barometer scanned more than a billion job ads across 27 countries and found a labour market splitting into two tracks. Roles PwC calls "professionalised", where AI takes the routine task off a person's plate and leaves the judgment to them, are growing twice as fast and paying 42% faster wage growth than roles it calls "democratised", where AI makes the job easy enough that the expertise stops mattering. Jobs that explicitly require AI skills are growing 69% against 9% for the market overall, and carry a 62% wage premium. The companies most able to use AI well are growing headcount faster (52% versus 36%) and paying more (24% versus 17%) than everyone else, and PwC's "super-star" tier posted 163% labour productivity gains.
Here is what matters for the entry-level argument specifically: PwC found AI-exposed entry-level roles that require senior-level skills, judgment, leadership, have grown 35% since 2019 (Read that again: Entry-level roles that require senior-level skills!). Every other kind of entry-level role has declined 10% over the same stretch. That is not "entry-level is dying." That is "entry-level is bifurcating," and most of the roles on the losing side of that split are the ones that used to be how you built a career.
Stanford's Digital Economy Lab put a sharper number on the losing side. Their "Canaries in the Coal Mine" research, led by Erik Brynjolfsson, tracks employment for workers 22 to 25 in occupations most exposed to AI. As of their June 2026 update, that cohort's employment sits 19% below where it would be if it had tracked their less-exposed peers, up from a 15% gap a year earlier. Experienced workers in the same occupations show no comparable gap. The mechanism is not mass layoffs. It is reduced hiring and attrition: companies just stop bringing juniors in. The effect is concentrated exactly where you would expect, occupations where AI substitutes for the task rather than assisting with it. Where AI complements the work instead of replacing it, employment is flat or rising, especially for the people who already have the judgment to direct it.
My Read: A Ramp, Then a Decline, Then a Hand-Over
Put those two studies together and you get the shape of where this goes. In the near term, the junior roles that survive will cluster around the gaps AI leaves: tuning the model's output, handling the exceptions it can't resolve, doing the physical tasks it can't touch yet. That is real work, but it is support work, not the apprenticeship work that used to teach someone how a business actually runs.
Medium and long term, I think the erosion keeps going. Companies will keep delivering more with less, and the people still employed will be asked to do more, mostly monitoring the agents, catching what they get wrong, feeding them the context they're missing. The humans left standing will be the ones who are genuinely good at managing AI output and who already know how the work is supposed to look when it's right, which is precisely the kind of judgment you used to build by doing the junior version of the job for a few years.
We are pulling up the ladder faster than we're building a new way to climb it.
We're also still early. Right now companies are in the ramp-up phase, figuring out what the technology can actually do and racing to implement it. That ramp doesn't last. As the tooling matures and organizations get past the experimentation phase, I expect the job market around AI itself to flip from growth to decline, handed off to a smaller, steadier support-and-maintenance function. The jobs created by AI's arrival are not all permanent. Some of them are scaffolding.
The Obvious Answer that is Still Missing from the Budget
One obvious fix here is training and worker development: take the judgment and institutional knowledge that used to get built through years of junior work, and build it deliberately instead, since the job market won't do it for you anymore. I don't think most organizations will spend the money. They should. They won't, for the same reason they never fully funded training and worker development when it was merely a good idea instead of a structural necessity: it shows up as a cost this quarter and a benefit three years out, and quarterly incentives win that fight almost every time. That is the actual risk sitting underneath the PwC numbers. Demand for senior people who can direct AI well is rising. The traditional pipeline that produced those people is shrinking. Something has to replace it, and right now nothing is.
The Wildcard
Robotics is the piece of this I watch and can't yet put a number on. It has been compounding quietly for decades without a breakthrough moment, and the signs suggest one is closer than it was, though I wouldn't claim a tipping point is visible yet. When it lands, it removes the one barrier that has kept AI's labour impact mostly confined to screens: the physical task. Pair language-model-grade judgment with robotic execution and the two-track split I'm describing above stops being about knowledge work and starts being about almost all work. That is a bigger conversation than this post, but it belongs in the long-term version of this picture.
A Warning Sign Hiding in the "AI User" Job Title
One more pattern worth mentioning: as AI matures, I expect roles built around "being an AI user" to disappear, not because AI failed, but because using it stops requiring a dedicated person. A role that exists to operate a chat window is a role built on the assumption that AI stays a tool you consult rather than a system that runs the workflow. I made a related point in The AI Savings Gap: drafting slides, summarizing a transcript, producing a first-pass model, was never the expensive part of the work. The value was always in the judgment layered on top. A role built to prompt the chatbot for admin tasks is a role built around the easy layer, and the easy layer is exactly what disappears first. When an organization still has headcount dedicated to "prompting the AI," that is usually a sign it is using AI as a chat-based assistant rather than as part of a true agentic workflow, and that gap is familiar. We saw the same pattern with the last wave of digital workflow automation: companies bought the tools, but struggled to actually change how work got done, because adopting the technology and changing the operating model are two different projects, and most organizations only ever fund the first one.
So, Now What?
Stop asking whether AI will take jobs. It already has, in a specific and measurable place: the routine entry-level work that used to double as on-the-job training. Start asking which track your organization's roles sit on. If a role is "professionalised," AI clearing the routine work and leaving judgment to the person, you are on the winning side of PwC's 42% wage-growth gap, and the job is to keep building that judgment deliberately since the market no longer builds it for you. If a role is "democratised," AI making the job easy enough that expertise stops mattering, you are on the side that's shrinking, and the honest move is to retrain toward the first track or plan for the headcount reduction rather than be surprised by it.
Either way, the organizations that come out ahead won't be the ones that used AI the most. They'll be the ones that kept training a pipeline of judgment while everyone else let the ladder erode. I don't think anyone has this fully solved yet, including me. Is the junior pipeline actually eroding on your teams, or is this overstated? What, if anything, are you doing about it? Tell me what you're seeing.
Comments
Post a Comment