When we picture an org chart (née organogram), we tend to picture a roughly pyramidal shape: wide at the bottom, tapering to a point at the top. Frederick Taylor codified this structure in 1911: in a Taylorist organization, managers plan, workers execute, and value flows upward through layers of oversight. This “pyramid schema” was designed for a world where information was scarce and coordination required humans at every rung of the ladder.

Fast forward a century and some change, and that world no longer exists. We’re living through the most consequential restructuring of corporate org design since the invention of middle management. The data is starting to reflect it. It’s clear enough that the pyramid is crumbling. The question is what will replace it, and whether companies figure that out intentionally, or stumble into it by accident.

The Hollow Middle

Let’s start with what’s happening on the ground.

At Moderna, HR and tech now live under the same roof, overseen by a single Chief People and Digital Officer. At one AI-first healthcare company, a team of ten software engineers was replaced with a three-person unit overseeing AI agents. At Amazon, layers of middle management are being stripped out as part of a broader push toward a leaner, AI-ready structure.1

Amazon’s move is worth dwelling on. CEO Andy Jassy mandated that each organization increase its ratio of individual contributors to managers by at least 15% by the end of Q1 2025, arguing that fewer managers would remove layers, flatten organizations, drive decision-making closer to the front lines, and decrease bureaucracy.2

If AI now handles data aggregation, status reporting, and routine decision-support, then the rationale for the middle management layer dissolves. Gartner predicts that through 2026, 20% of organizations will use AI to flatten their organizational structure, eliminating more than half of current middle management positions.3 Meanwhile, among organizations with extensive agentic AI adoption, 45% already expect reductions in middle management layers within three years.

The Illusion of Flatness

Here’s where I want to push back on the prevailing narrative, though. The popular shorthand is that AI creates “flatter,” less hierarchical organizations. What’s actually happening is a different form of shapeshifting.

As we’ve established, most orgs today look like a pyramid: small leadership at the top, a middle tier of managers and specialists, and a large base of entry-level workers on routine tasks. As AI agents absorb those entry-level tasks (e.g. data gathering, processing, and reporting), companies are hollowing out the base while repurposing the middle to train and oversee agents. The result is often a diamond shape, as PwC observed in a recent deep dive.4

But PwC’s analysts caution against the diamond, instead arguing for the inverse: an hourglass shape: wide at the top, narrow in the middle, and wide again at the base, which consists of a new generation of AI-literate generalists who can ramp up faster and contribute at higher levels earlier. Within that hourglass, the unit of execution is the pod — a small, cross-functional team with flat internal hierarchy, mixed human-agent composition, and broad authority over how it achieves its mandate.

Retaining the wide base is essential. If you hollow out entry-level roles entirely, then you’ve sawed off the apprenticeship ladder that produces future senior talent. The diamond may look efficient today, but it precipitates a leadership crisis in a decade, a corporate equivalent of population collapse.

This is one of the defining org design questions of the next decade, and most companies aren’t yet asking it deliberately.

Agents and the Accountability Gap

The current AI era isn’t about rote automation. It’s agentic AI that’s all the rage: systems that don’t just answer questions when prompted, but plan, act, and adapt autonomously across multi-step workflows. Gartner predicts that 15% of day-to-day work decisions will be made autonomously through agentic AI by 2028, up from essentially none in 2024.5

This breaks the top-down hierarchical model, which presupposes human nodes. If agents coordinate workflows, traditional managerial spans of control increase and the number of hierarchical layers decreases, with human managers increasingly responsible for orchestrating hybrid teams of humans and agents.6 That hybrid orchestration may well be the new core managerial skill set. Alas, almost nobody is trained for it. 53% of people managers are concerned they may not be good at supervising AI-augmented teams, and 63% of non-managers are now hesitant to pursue management roles due to concerns about leading AI-augmented teams.7

But there’s a deeper problem underneath the skills gap: accountability. Traditional hierarchies are inefficient, but they do at least one thing well: every decision has a human name attached to it. The org chart is, among other things, a liability index.

Agentic AI breaks that. When an agent autonomously handles a pricing call, a customer resolution, or a procurement decision, the choice didn’t come from a person. It came from a system designed by people, sure, but the discrete decision was made by the machine. In an org chart where individual contributors are now “orchestrating” agents without traditional oversight, responsibility lands nowhere cleanly. Executives make the structural bet (”we’ll use agents for X workflow”) but are often several layers removed from what the agent is actually doing day-to-day. And as agentic AI moves into consequential territory (think performance evaluations, credit decisions, or hiring calls) the question of which human is accountable becomes a legal exposure, not just a matter of managerial philosophy.

The pitch for hourglass-shaped organizations is that decision-making moves to the edges. But in an agentic org, the edges are increasingly occupied by AI systems. You’re distributing execution without distributing accountability. That’s a recipe for diffusion of responsibility at scale, and it’s a risk few are ready for.

There’s a subtler risk too: because AI systems are trained on existing data and optimized for measurable accuracy, they tend toward the statistically average outcome, which isn’t always the right one for a specific customer or context. Organizations need to deliberately instrument their processes so that human managers can actually inspect what’s happening at the edges, not just whether the workflow completed or not.

Toward Goal-Driven Hierarchies

The future organogram isn’t primarily a chart of persons. It’s a chart of goals and accountabilities, with humans and agents assigned to them fluidly. This reframing matters. Most companies are trying to insert AI into their existing human structure, as tools, assistants, and augmentation. That’s an imperfect frame. The companies getting ahead are redesigning around outcomes, then asking: what combination of humans and agents best achieves this? Who owns the accountability?

A side effect of AI-driven process design is that individual contributors now need skills traditionally associated with management, such as delegation, clear scope of work, budget allocation, and objectives to guide AI systems.8 Everyone is becoming a manager of something. Just not necessarily of other humans.

A few practical implications, if you’re running a team:

• Stop managing tasks; start managing outcomes. If your management layer exists primarily to track work and report upward, it’s on borrowed time. Redesign around accountability for results, not supervision of process.

• Rethink your entry-level pipeline deliberately. The diamond model might look efficient today and cost you dearly in five years. If your organization requires institutional knowledge and judgment to scale, you need to think about how people develop those aptitudes. AI-replacing the entry rung eliminates that developmental path.

• Treat agentic AI like a headcount decision, not a software decision. When you deploy an agent to handle a workflow, you’re making a workforce composition choice. It should go through the same scrutiny as any hire. What is this agent accountable for? Who owns it? How do we evaluate performance? Most critically: when it gets something wrong, whose name is on it?

The pyramid had a great run, from ancient Djoser to the present-day Fortune 500. But it was always a coordination system, not an accurate picture of how value is created. AI appears poised to be the forcing function that finally lets us design organizations around how value actually flows — horizontally, networked, outcome-driven — rather than how authority has historically asserted itself.

The org chart isn’t dying. It’s simply evolving. And as time ticks down on our old ways of working, the hourglass is a fitting emblem for a new era.

Special thanks to Rob Lentz, Veeral Shah, and Vikas Mehta for workshopping some of the ideas in this post.

1 Fortune

2 Databricks

3 Gloat

4 PwC

5 Deloitte Insights

6 MIT Sloan Management Review

7 EY

8 Databricks