written by: Lovro Mršić
This summer, while walking along the promenade in several places along the coast, I noticed something I hadn’t seen in previous years. Almost every stand, whether renting pedal boats or selling kebabs, welcomed the season with a new large banner and a neat, seemingly solid logo, which until recently was only found at the hotel across the street.
For years, a board with the handwritten “Bakery Pero” was sufficient for the same job, so the question naturally arose: where did everyone suddenly get a good logo? Either the budgets of small businesses increased overnight, or the cost of production fell so much that the budget no longer mattered. We all know the answer, and that’s why I find the other question more interesting: why should this little detail on the promenade concern us?
To answer it, we first need to see how it came to be. Generative artificial intelligence is the fastest adopted technology in history: according to the Stanford AI Index for 2026, within three years of its release, more than half of adults had used it, significantly faster than the internet or personal computers spread at the same stage.
Stand owners are part of that statistic just like all of us, which means they have, like all of us, seen what the tool can do, added two and two, and concluded that a generated logo would be great for their business. The trouble is that a good portion of them added the same numbers, resulting in hundreds of logos that are technically correct, almost free, and so similar that, in a way, they repel more than they attract.
The First Step
However, this uniformity is not a problem in itself, but a symptom of something that happened a step earlier. Because that logo was not created by an agency before; it was made by someone’s nephew in their third year of design, a student alongside their studies, or a junior in a small agency who was given “that small job for a trial”.
It was too small for anyone to call it a job, but it was precisely on such jobs that things were learned that cannot be learned elsewhere: what a client looks like, what a deadline looks like, what a bad brief looks like, and what two hundred identical logos look like when you see them side by side. Today, that same logo is made by the owner themselves, in two minutes between two customers, and no one has been fired in the process. Simply put, no one is calling the student anymore.
That is what the promenade actually shows. The first step is not disappearing from the top, with layoffs and headlines in the newspapers, but from the bottom, quietly, in jobs that no one has even counted. And the mechanism that made the logo free does not stop at logos. The same tool, for the same reasons, is making everything that until recently was called junior work free, and with it disappears the only known way to transition from junior to senior.
So let’s go in order. An internship, in theory, is precisely that bridge: a place where someone who has a degree but no experience does real work for the first time alongside someone who knows how to do it. In practice, for most students and recent graduates, it looks like this today. First, several hundred applications are sent, largely to ads that never receive a response because the competition is such that the employer has no reason to reply.
The numbers confirm this: according to Handshake, the number of entry-level ads has fallen by 15 to 16 percent annually, while the number of applications per ad has increased by 26 to 30 percent, and in the tech sector, the number of internships has fallen by 30 percent since 2023, alongside a simultaneous increase in applications. When you factor in that approximately 35 percent of ads labeled as entry-level require three or more years of experience, you get a market where the first job has become something that is won in a lottery, not through qualifications.
Those who win that lottery, in most cases, do not get what they were looking for. They get a role that no one designed because the company did not take an intern because it had work for them, but because it had a program, a quota, or goodwill. There are no mentors who have time, no defined tasks, no end where someone sits down and says what was good and what was not. There is copying, organizing spreadsheets, and waiting, and after three months, there is a line in the resume that looks like experience, but it is not. Exceptions exist, and we mostly know them by name precisely because they are exceptions. When this is mentioned to someone, the response is almost always the same: “I also brewed coffee for six months.” That is commendable, but that was a different economy, where after coffee came a job.
It would be easy to conclude that the problem lies with companies that do not care, but that would be incorrect because most people who run these internships genuinely believe in young people and sincerely wish that the internship would be better. Internships are not poorly designed because someone does not care; they are poorly designed because good design would cost someone, and that someone has one open position, one deadline, and, since there is a tool that can do the job already this afternoon, a very good reason not to take an intern at all.
That reason is neither malice nor shortsightedness, but calculation, and it is worth writing it out to the end because it is rarely spoken aloud. A junior costs a full salary, and along with it, a part of the work week of some senior who introduces them to the job, corrects and explains, and this lasts about a year before they become net positive for the team. When they finally become net positive, they are also the most mobile person in the company, so there is a real probability that they will leave for a thirty percent higher salary somewhere where they did not have to be taught.
The artificial intelligence tool, on the other hand, costs less than coffee for the team and works already this afternoon. Managers know this: in a survey conducted among them in 2024, 70 percent said that AI can do the job of an intern, and 57 percent trust the results of the tool more than the work of interns or recent graduates. If you have one open position, one deadline, and one budget, skipping the junior is not a lazy decision. It is the right decision.
Data shows that it is being made, and precisely where we would expect it. A Harvard study found that in companies that adopted generative AI, entry-level hiring fell by about 80 percent per quarter, while senior employment in those same companies continued to grow. A Stanford analysis of ADP salary data shows the same from the other side: employment of 22 to 25-year-olds in jobs most exposed to AI is falling by 3.8 percent annually, while employment of 35 to 40-year-olds in those same jobs is rising. In other words, AI is not erasing professions, but erasing their lower tier. The job remains, it just can no longer be entered.
However, we must be fair to the other side because the evidence is not unequivocal. A Fed study based on data from over a million companies found no link between AI adoption and a decrease in the number of ads, and Indeed describes what is happening as “experience creep”, a trend where more and more experience is required for entry-level positions, which predates ChatGPT. But for the question that concerns us here, that debate is not actually decisive. Whether the cause is the tool, interest rates, or employer habits, the mechanism is the same: every individual decision to skip a junior is reasonable, but their sum is not.
Because here lies the real catch, and that is why the problem cannot be solved company by company. A company that trains juniors is training them, on average, for a competitor that did not do so. It pays for a year of learning, the competitor pays thirty percent higher salary and gets a ready-made person. Rationally, no one should go first, and there is no one to blame for that; it is a textbook coordination problem, the same kind of problem that prevents anyone from building a road alone even though everyone wants a road. And coordination problems have one unpleasant characteristic: the market does not solve them on its own, but deepens them.
And all of this still rests on one assumption that no one has verified: that the tool is indeed cheaper than coffee.
The Price That Has Never Been Real
That assumption, in fact, does not hold. In a text about the AI bubble, I wrote a year ago about who pays for this race: about the capital that circulates within the same ecosystem and about the company that loses money even on a subscription of 200 dollars a month because inference costs are rising faster than revenue. What I did not fully say then is that this loss must be accounted for somewhere, and it is accounted for, among other things, in hiring decisions. Because when a manager calculates that the tool costs less than coffee, they are calculating with a price that someone else subsidized to gain market share.
The bill is now coming due. Gartner published a forecast in June of this year that by 2028, the costs of AI coding will exceed the average salary of developers due to rising token consumption and the industry’s shift to pay-per-use instead of fixed subscriptions. The jump is already visible: bills for AI tools per developer are rising from 20 or 100 dollars a month to 2000, even 5000 dollars. In other words, a generation has been cut from the labor market based on a price that has never been real, but was a result of venture capital subsidizing the calculation.
Part of this is already being corrected, especially among those who were the first to go in the opposite direction. IBM, whose CEO publicly announced in 2023 that AI would replace 7,800 jobs, is tripling entry-level hiring in the U.S. this year, and according to their HR director, “for software developers and all those jobs that they tell us AI can do”.
Her explanation sounds like it was written for this text: if we do not invest in entry-level, what happens in three to five years? There is no pipeline, the source simply dries up. IBM is not alone: a survey by Orgvue among more than a thousand business leaders showed that 55 percent of those who laid off due to AI now admit they were wrong, and Forrester predicts that half of the layoffs attributed to AI will be quietly reversed by the end of 2026.
But here lies a difference that is easy to overlook. Layoffs can be reversed, a person can be brought back or a new one hired. Three years in which no one was trained cannot be reversed by anything because a senior who was refused training cannot be hired. The junior you skipped in 2024 is not waiting in a drawer; in the meantime, they either went into another profession or spent three years doing something from which they learned nothing. The gap in the pipeline does not close when the decision changes. It travels through time and arrives when it is most expensive.
Who Sits at the Table
There is also one reason that has nothing to do with costs, and it should be the most convincing for companies. Most of them today want to sell to Generation Z and, increasingly, Generation Alpha, and they are trying to do so without a single representative of those generations in the rooms where decisions are made. It is not about symbolism. It is about the fact that you will do a product, campaign, or application for twenty-year-olds better if you have a twenty-year-old on the team who can tell you that what you have devised will not be opened by anyone of theirs.
And precisely here, AI, instead of being a reason for not hiring, can be a reason for the opposite: if the tool takes over copying, spreadsheets, and first drafts, the junior no longer has to go through six months of brewing coffee to get to a job where they are useful. They can work from day one on what makes it worth having them – bringing context, language, and market instinct that seniors only know from research. IBM has, after all, reshaped its entry-level roles precisely this way, with less routine coding and more client work. Structured internships in the era of AI are not more expensive than unstructured ones; for the first time, they are actually possible.
That is why I ultimately return to the sentence that has long become a cliché: we believe in young people, they are the future. It sounds generous, but it is not an obligation, but a description of something that will happen anyway. We are the future, whether you believe in us or not. What is not indifferent is whether we will enter that future as people who have learned the job or as people who have spent three years sending emails. “We believe in young people” is said in public. “Juniors are not worth it” is said in budget meetings. Both sentences are true, and only one of them decides what really happens. And until that changes, the logo on the promenade will remain what it is: technically correct, almost free, and the same as all the others.
