The traditional corporate pyramid is disappearing. For decades, companies hired large numbers of high school and college graduates at the bottom. Some left, some plateaued and some advanced. That broad base supplied the progressively smaller layers of experienced employees, managers and leaders above it. Hence the pyramid.

Now corporations are shrinking the base. Pave, a compensation benchmarking and workforce analytics company, analyzed more than 8,700 companies and found that entry-level employees declined from 6.8% to 4.6% of their workforces over two years, a 32% drop. Stanford researchers using payroll data covering millions of U.S. workers found an even sharper pattern in AI-exposed work. Employment among 22- to 25-year-olds in highly AI-exposed occupations is now 19% below where it would have been had it kept pace with young workers in less-exposed occupations.

HR Executive calls the resulting organizational structure a “career diamond”: fewer employees at the bottom, a large population of experienced workers in the middle and fewer leaders at the top. But that shape can only exist temporarily.

Companies may have a large middle today because those employees entered the workforce when the bottom was larger. For a while, companies can also hire experienced mid-level people developed elsewhere. But if everyone shrinks the bottom, there will eventually be fewer experienced workers to hire. The middle cannot remain large if the population feeding it is small. In other words, companies still must grow experienced talent. They will simply have far fewer people from whom to grow it.

That changes the risk. Imagine a company that once hired 100 entry-level employees. Some would leave, some would fail to develop, and others would eventually become the experienced employees and leaders the organization needed. If one person failed to make that progression, the company still had 99% of its original pool from which to grow future talent.

Now imagine the company hires only 20. It begins with just 20% of its former pipeline capacity. Lose one of those employees and only 19% remains. The same single failure that once left 99 potential future employees in the pipeline now leaves 19.

That is one risk created by shrinking the bottom. Companies are dramatically reducing their overall capacity to absorb developmental failures. Reducing entry-level headcount may lower labor costs, particularly when some of the work is replaced by AI. But it also concentrates talent risk. Every poor hiring decision, bad manager, regrettable departure and failure to develop someone now removes a larger percentage of the company’s potential future talent.

There is a second problem especially in the world of AI. The smaller group of entry-level employees has fewer opportunities to do the work through which previous generations built judgment, risk tolerance and experience.

Old-school entry-level work served two purposes. It got relatively routine work done, and it developed the people doing it. A junior analyst prepared the first analysis and someone more experienced reviewed it. A new software engineer handled routine coding and debugging. A young employee sat in on customer calls, drafted communications, made low-consequence decisions, got things wrong, received feedback and tried again.

The work provided repetition with relatively limited risk. Over time, employees were given more complicated problems and greater responsibility. That is how they developed judgment. AI is absorbing exactly this kind of work.

Companies are hiring fewer beginners while simultaneously removing some of the experiences that turn beginners into experts. Neither training nor tenure alone can fill that hole. Classroom learning about risk taking cannot substitute for acting without complete information. Understanding adaptability does not teach someone what to do when circumstances change. Judgment develops through exposure, repetition, feedback and progressively more difficult decisions.

If companies want to operate with a much smaller entry-level population, the risk/reward equation changes. They have to protect and develop the people they do hire. They need those employees to stay, and they need to recreate the developmental mechanism that repititious junior work once provided.

The model already exists its white collar apprenticeship. For centuries, apprenticeship was a primary way skilled workers were developed. An apprentice worked with a master-craftsman for a defined period, learning the trade while doing real work. The apprentice observed, attempted the work, received correction and gradually developed the capability to work independently.

That is strikingly close to what knowledge work will require if we change the pyramid to a cylinder.

If the old process was effectively “hire a large group, let them learn through doing, and advance the ones who excel,” a smaller entry-level population requires something much more deliberate. It requires dyads or small groups of junior employees working closely with seasoned people who are expected to transfer judgment.

Pushing this to L&D will not accomplish what is needed. L&D can teach frameworks and foundational workplace skills. It cannot manufacture the repeated experience through which judgment develops. That has to happen in the work.

If a junior employee no longer prepares the routine analysis because AI produced it, a senior employee must coach.  They could require the junior employee to reach a conclusion before reviewing the AI output.

They would bring junior employees into difficult customer conversations, negotiations, escalations and decision meetings earlier and they could do more than have them observe. Seniors could ask juniors what they noticed, what they would have done and why the experienced employee might have made a different choice.

In other words, the modern-day craftsman would give the apprentices consequential work with guardrails. Let them make the recommendation before a senior person approves it. Let them lead the conversation with someone experienced present. Give them decisions where mistakes have consequences but can still be caught and corrected. Experienced employees would make visible the judgment that has become automatic to them..

In other words: Observe. Attempt. Review. Correct. Repeat. Increase the difficulty.

This is not a new development theory. It is an old model becoming relevant to a new kind of work. Britain is already expanding white-collar apprenticeships in law, finance and technology. PwC, Barclays and Freshfields are among the employers offering routes that combine paid work, formal learning and direct exposure to experienced professionals.

IBM is approaching the same problem from another direction. Rather than sharply reducing its entry-level population, it has said it plans to triple U.S. entry-level hiring while redesigning those jobs around work that remains valuable in an AI-enabled organization. Entry-level roles are shifting away from routine execution toward analysis, problem solving, collaboration, client interaction and AI oversight.

These approaches are not mutually exclusive. Companies can redesign entry-level jobs and continue hiring substantial numbers of beginners. Or they can decide that AI allows them to hire far fewer and build a white-collar apprenticeship model around the people who remain.

What they cannot do indefinitely is continue drawing experienced talent from a middle that was created by yesterday’s much larger entry-level population.

________________________________________________________________________________________

Sources

Brynjolfsson, E., Chandar, B., & Chen, R. (2026, August 12). Canaries in the coal mine? Six facts about the recent employment effects of artificial intelligence. Stanford Digital Economy Lab.

Colletta, J. (2026, August 21). The “career diamond”: HR’s next competitive liability. HR Executive.

Epstein, S. R. (1998). Craft guilds, apprenticeship, and technological change in preindustrial Europe. The Journal of Economic History, 58(3), 684–713.

Humphries, J. (2010). Apprenticeship. In Childhood and Child Labour in the British Industrial Revolution. Cambridge University Press.

Knuth, C. (2026, March 23). How AI pressure creates diamond-shaped org structures. Pave.

McConnon, A. (2026, March 2). The bottom rung returns as AI reshapes entry-level jobs. IBM Think.

Mokyr, J. (2019). The economics of apprenticeship. In M. Prak & P. Wallis (Eds.), Apprenticeship in Early Modern Europe. Cambridge University Press.

The Wall Street Journal. (2026, August 22). British kids are skipping top colleges for white-collar apprenticeships.

Wallis, P. (2008). Apprenticeship and training in premodern England. The Journal of Economic History, 68(3), 832–861.