From Pyramid to Diamond: Why the Shape of Your Organization Is Creating a Judgment Crisis

For decades, the organizational chart in business looked like a pyramid. Wide at the base, narrow at the top. Large numbers of junior workers doing foundational, analytical and often repetitive work at the bottom, a smaller layer of experienced managers in the middle, and a narrow band of senior leaders at the top. The shape was not accidental. It reflected how knowledge work actually develops people: you start at the bottom, you do the hard work, you make mistakes in consequential enough ways to learn from them, and over time, if you grow, you move up.

In recent years and months, that pyramid has been changing shape. And the implications for the quality of human judgment inside organizations may be more serious than most leaders have begun to reckon with.

The Pyramid Becomes a Diamond

AI adoption is compressing the base of the organizational pyramid. The analytical, iterative, transactional work that used to fill early careers, the work that was slow, messy, and occasionally humbling, is being absorbed by AI platforms that do it faster, more consistently, and at a fraction of the cost. As a result, organizations need fewer people at the entry level. The base narrows. The middle begins to grow as practitioners advance faster than in a previous era. And the top remains largely intact.

In the United States, PwC has confirmed plans to reduce entry-level hiring by roughly one-third over the next three years. Its AI Assurance Leader, Jennifer Kosar, has argued that new hires will increasingly enter the workforce as "reviewers and supervisors" as AI absorbs many routine tasks traditionally performed by junior staff. Recent labor-market research likewise suggests that firms adopting AI are reducing demand for early-career roles, with studies reporting meaningful declines in employment among younger workers in AI-exposed occupations. In the UK, graduate hiring has already contracted across the Big Four, with KPMG reducing intake by 29%, Deloitte by 18%, and EY by 11%. The Stanford AI Index identifies finance, law, accounting, and professional services among the occupations most exposed to AI-driven task automation and augmentation. The ICAEW has described the emerging organizational shape as a "diamond model": a narrower base, a broader middle of technical and managerial practitioners, and continued reliance on human judgment where AI remains insufficient. The shape is increasingly visible. What is less understood is how organizations will develop the experienced middle layer that model depends upon if fewer people enter through the base.

On a simple org chart, that might look like progress. Leaner at the bottom, more experienced in the middle, efficient at the top. But the diamond shape carries a structural problem that does not appear on any chart: the middle layer, the broad band of practitioners now carrying the weight of the organization's knowledge work, was never given the developmental conditions that built the capability the role assumes they have.

The judgment gap begins at the bottom. It forms quietly, in the absence of the developmental pressure that early-career work used to provide, before anyone has a reason to look for it. But it does not become visible there. It surfaces in the middle, precisely where the organization is now depending on judgment most, and precisely where there is the least tolerance for discovering it is not there.

How the Diamond Loses Value

Think about how a diamond actually forms. It develops under enormous pressure, over a long period of time, in conditions that force out impurities and create a structure of strength and clarity. The pressure is directly responsible for the diamond's value. A diamond formed without sufficient pressure, or formed too quickly, develops inclusions: internal flaws, cloudiness and structural weaknesses that compromise its clarity and reduce its value. A stone with too many inclusions is almost worthless, regardless of its size or shape.

The same principle applies to professional judgment. Judgment does not develop in classrooms. It does not develop through instruction or exposure to well-constructed frameworks. It develops under pressure, in the context of real work, when the stakes are real, when the consequences of getting it wrong are felt, and when there is structured reflection afterward to convert the experience into something a practitioner can actually use.

Here is what matters about the timing. Inclusions do not form after a diamond is complete. They form during growth, when the conditions are disrupted, when the process is rushed, when foreign materials enter the crystal structure before it has fully set. The disruption is invisible from the outside. The stone continues to grow, continues to look like a diamond, and arrives at its full size carrying flaws that were locked in early and cannot be corrected later. By the time the diamond reaches full size, whatever formed inside it already has. The same is true of professional judgment. The early-career work that AI is now absorbing was not valuable because it produced good output. It was valuable because it created the sustained, uninterrupted developmental pressure through which judgment formed. Compress that process, skip the difficult early stages, and the practitioner who arrives in the middle of the diamond looks complete. They perform well on available metrics. But they carry inclusions that do not show up until the moment the work requires something AI cannot produce.

That moment arrives precisely in the middle of the diamond. When the broad layer of experienced practitioners is asked to evaluate what AI is generating, to catch what the model missed, to make the call in an ambiguous situation where the data points in two directions at once, the inclusions become visible. The middle goes soft. And the organization that built its structure around a diamond shape, without understanding what it takes to form a flawless one, discovers a quality problem hidden within its efficiency story.

The Data That Looks Like Good News

Some will argue that the data tells a different story. That AI is not stunting development but accelerating it, placing junior practitioners into more judgment-intensive work sooner and building competence faster than the old pyramid allowed. It is a reasonable argument on its face. It is also worth examining more carefully.

Consider what AI-assisted diamond production reveals about accelerated growth. One common laboratory method, Chemical Vapor Deposition, grows diamonds quickly in controlled conditions. The stones look excellent. The metrics are strong. But CVD diamonds frequently require post-growth heat treatment to achieve their color grade, because the accelerated process alone cannot produce the result that sustained natural formation does. Without that corrective intervention, what you have is a stone that is structurally present and metrically plausible, but not what it appears to be.

The same dynamic is playing out inside AI-augmented organizations, and the PCAOB inspection data makes it visible. Big Four audit deficiency rates fell from 26 percent in 2023 to 20 percent in 2024, a meaningful improvement that regulators have acknowledged. AI tools have been central to it. Expanded coverage, better sampling, tighter documentation, faster anomaly detection: these are areas where AI performs exceptionally well, and where it has driven measurable gains. The efficiency story is real. And it is producing output that looks like quality.

But look at what the improvement has left behind. The deficiencies that AI is resolving are largely procedural and evidence-gathering failures, the kind rooted in declarative knowledge and systematic coverage. What remains, and what regulators continue to flag with persistence, are the judgment-intensive failures: testing estimates, evaluating management review controls, exercising professional skepticism in ambiguous situations where the data does not point cleanly in any direction. The PCAOB and the SEC have repeatedly identified professional judgment and skepticism as the source of the profession's most stubborn and recurring deficiencies. These are not problems that more coverage or better documentation will solve. They are problems that require practitioners who have formed genuine judgment, and that formation requires conditions that the AI-augmented early career is no longer providing.

The Judgment Debt No One Is Accounting For

What organizations are building, without recognizing it, is judgment debt. Like financial debt, it does not appear on any balance sheet. It accumulates quietly while the efficiency metrics look strong and the deficiency rates fall. And like financial debt, it compounds. Every cohort of early-career practitioners that moves through a system stripped of developmentally dense work adds to the balance. The payment comes due when those practitioners reach the middle layer and are asked to do the one thing AI cannot do for them: exercise judgment in conditions that are genuinely ambiguous, genuinely consequential, and genuinely without a clear procedural answer.

The current improvement in audit quality is most likely a leading indicator, not a destination. The deficiencies AI is eliminating are the easy ones, the ones that were always rooted in process, coverage and documentation. What is left is almost exclusively judgment-linked. And the pipeline that used to produce the judgment required to address those remaining deficiencies is narrowing at precisely the moment when its output will be needed most.

Organizations that fail to account for this debt will likely discover it the hard way. If the practitioners entering middle-layer roles over the next two to three years have moved through early careers that were systematically stripped of the developmental pressure that forms judgment, the deficiency rates that look so encouraging today could reverse sharply. Not because AI failed, but because the judgment required to supervise it, interrogate it, and catch what it missed was never formed in the people now responsible for doing exactly that. The data will confirm the problem at precisely the moment when the window to prevent it has already closed.

Recreating the Pressure

The answer is not to slow down AI adoption. The competitive pressure is real, the efficiency gains are real, and no organization is going to abandon tools that are genuinely making their work better and faster. The answer is to understand that the pyramid worked as a developmental system because the pressure was built in. The work itself created the conditions for judgment to form. The diamond, left to its own devices, does not.

Organizations that move to a diamond shape without deliberately engineering the developmental pressure that the pyramid used to create naturally will produce practitioners with inclusions: capable within defined parameters, brittle outside of them, and increasingly exposed as AI absorption moves more of the structured work away from human hands. By the time those inclusions are visible, the practitioners who carry them are already in the roles the organization depends on most.

What that engineering looks like in practice is a design question, not a training question. It means asking where, in the new organizational structure, the cognitive friction lies. Where are the real stakes? Where does a practitioner encounter genuine ambiguity, bear the consequences of a judgment call, and have the opportunity to reflect on what happened? If those conditions do not exist naturally in the AI-augmented workflow, they need to be created deliberately, through simulation, through structured stretch assignments, through developmental experiences that restore the pressure the diamond's shape has removed.

A flawless diamond is not formed quickly or easily. It takes time, the right conditions, and sustained pressure applied in the right way. The organizations that understand this and build the developmental infrastructure to recreate those conditions in an AI-augmented world will produce practitioners of genuine quality. The ones that do not will keep discovering inclusions at the worst possible moment, when the pressure is real, the stakes are high, and there is no simulation to fall back on.

The shape of the organization has changed. The requirements for what makes a practitioner truly capable have not.

---

Larry Durham is the President of St. Charles Consulting Group and co-author of The Talent-Fueled Enterprise: A Powerful Approach to Build Tomorrow’s Workforce and The Judgment Void: How AI is Dismantling the One Human Capability it Cannot Replace. With over 30 years of experience in organizational learning and talent transformation, Larry partners with Fortune 1000 companies to help them reimagine how skills fuel business strategy and performance. He is a recognized thought leader in skills-based talent management, learning transformation, and workforce agility—guiding organizations as they shift from static job structures to dynamic, skills-driven ecosystems. Larry’s work bridges strategy and execution, helping leaders connect skills data, technology, and operating models to create resilient, future-ready workforces.