AI Is Killing the Century-Old Organizational Pyramid

By Anish Batlaw
Global Head of Talent

The organizational pyramid — the hierarchy of specialized functions, coordinated through layers of management, that has dominated virtually every large institution for a century — is on its way out. And the cause of death is artificial intelligence.

Every era of growth over the last hundred years reinforced the pyramid as the best way to structure an organization. No company made that case more clearly than Ford, where, in the early 20th century, a pyramid of supervisors, foremen, and functional departments kept thousands of workers moving in concert on the assembly line. Between 1904 and 1914, as Ford scaled that structure, revenue per employee grew from $3,880 to $9,277.¹

Alfred Sloan took the pyramid further at General Motors. His divisional structure unlocked the ability to coordinate across multiple brands, geographies, and functions that no single manager could hold — and by 1955, GM was generating $19,940 in revenue per employee.² IBM pushed that figure to $123,432 by 1985³. Amazon hit $925,900 by 2008 ($1.45 million in 2026 dollars).⁴

The pyramid worked because it solved the central problem of modern business: how to organize large numbers of people toward a common output. It was so powerful that in the early twentieth century, an entire class of businesses came to be identified by their newfound ability to do it. They were called “organizations.”

Now AI is breaking that structure by expanding what any one person, or any small team, can accomplish. As AI agents absorb more of the analytical and executional work, the need for layers of management falls away — and with it, the pyramid itself.

What’s emerging in its place looks less like a flatter pyramid and more like a shape-shifting organism. In this new model, small, autonomous teams — what Oliver Dlouhy, founder and CEO of Kiwi.com, calls “Mission-Aligned Teams” — come together for an assignment, direct a network of AI agents toward a discrete outcome, and dissolve when the project is done. The teams coordinate laterally rather than reporting vertically. They set outcomes, define constraints, and make the judgment calls that require human context. In the most advanced cases, two to five people supervise fifty to a hundred specialized agents across customer operations, product development, financial reporting, and market expansion.

NVIDIA now generates $5 million per employee.⁵ Cursor, an AI coding startup, produces about $13 million. Revenue-per-employee figures like these have no precedent in the history of business. My belief is that, as they adopt AI and lose the pyramid, more companies will see similar jumps.

The implications for how companies are structured and who they hire (and fire) are significant. The organizational pyramid assumes strategy lives at the top, execution at the bottom, and that a whole layer of people exists to translate between them. That assumption produced one of the most repeated rules in management theory, that no manager should have more than six to eight direct reports. No more.

Meta’s new AI team, which is focused on superintelligence, operates at a 50-to-1 employee-to-manager ratio, while Mark Zuckerberg directly manages 25 to 30 people.⁶ At NVIDIA, Jensen Huang has roughly 60 direct reports.⁷

When companies widen spans and thin layers, they are making a statement about where thinking should happen in an organization. (They also tend to treat the cost savings as a byproduct, not the main goal.) Now that nearly every person has access to the same real-time data, the same AI tools, and the same capacity to build, everyone, regardless of where they sit, is expected to be a thinker.

Exhibit 1: NVIDIA broke the ceiling. AI-native upstarts already exceed it by an order of magnitude.⁸

Annual revenue per employee. Open circle = FY2020. Filled circle = FY2025. Six U.S. tech mega-caps on a log scale, with private AI-native benchmarks plotted as stars overhead.

It’s true, the companies I’ve mentioned aren’t representative of all companies. Most firms aren’t unicorns with singular founders operating at the frontier of artificial intelligence. So it’s fair to ask, “Even if the pyramid is dead at AI rocket-ships like Cursor or among the Magnificent Seven, won’t the vast majority of firms still require work to take its typical triangular shape?” No, not necessarily.

Kiwi.com is not a hyperscaler or AI-native startup. It’s a mid-market European travel technology platform in a margin-compressed, highly competitive industry. But over the past eighteen months, Oliver and his team have systematically dismantled Kiwi’s organizational pyramid into Mission-Aligned Teams — all built around artificial intelligence.

There have been measurable results. The customer service function, rebuilt around an AI agent named Nathan, now consistently outperforms the human teams it replaced by 15–20% on customer satisfaction. Engineering throughout, measured by merge requests per developer, nearly doubled even as total headcount fell sharply. Business intelligence that previously required a team of analysts and a multi-week backlog is now available to everyone in the organization in real time, through an agent named Stephen, accessible by tagging it on Slack.

Cost, as you might expect, went down. But revenue, which you might not expect, went up 35% year-over-year. The AI efficiency hacks had freed up the money and the people necessary to try new things. Budget airlines, which were previously not worth integrating because the cost of maintaining them on the website outweighed the revenue they generated, are now part of their offering. Kiwi now carries more airlines than at any point in its history. A market that did not exist eighteen months ago now contributes meaningfully to its revenue. The pyramid had been more than expensive. It was limiting what the business could attempt.

Most CEOs I know understand this is true for their firms, too. And yet they’re often unsure of how to integrate AI. Last month, I sat in a closed-door room with about fifteen public-company CEOs, board directors, and C-suite executives — none in our portfolio — to talk about AI. What I heard was paralysis. They told me they wanted to act but were not sure where to begin, so they had asked IT to come up with a strategy. The board members told me they had charged their CEOs to do the same. No one was taking ownership. Everyone was passing the buck.

When they’re unsure of what to do, companies default to restructuring. They remove layers, reduce headcount, automate the obvious. But the right frame for this moment is redesign. Elon Musk has articulated a principle that applies directly to this moment: delete before you optimize. If a process, a layer, or a requirement cannot be justified from first principles, remove it. The instinct to improve what exists is almost always wrong if what exists should not exist.

Companies should start with a simple question: If we were building this company today, with the tools now available, what would we actually build?

As growth investors, that question is central to how we think about every company we back. When I work with CEOs migrating their business toward a new structure, there are five things I always tell them.

1. Look honestly at what your middle layers actually do all day.
Kiwi found that a substantial portion of analyst and coordinator time was devoted to tasks that AI agents now handle in minutes.

2. Identify one end-to-end process and rebuild it from scratch as AI-first.
Reintroduce humans only where judgment, relationship, or accountability genuinely requires them.

3. Make context documentation a priority.
In most organizations, AI underperforms for one reason: context. The capability is there. The institutional knowledge to direct it is not. Agents operating on well-structured institutional knowledge perform dramatically better than those working on incomplete information. The single biggest constraint on moving faster, as Oliver from Kiwi puts it, is context still residing in human minds rather than in systems where agents can act on it. Making what your organization knows machine-readable sits on the CEO’s agenda, not the CIO’s.

4. Start with yourself.
Most CEOs I know are focused on staying essential — on being the person the organization can’t function without. Contrast that with Oliver at Kiwi, who spends his weekends designing workflows around his own role and working with his platform team to build the agents that make him more productive. His mission, he says, is to make himself redundant. Hiring a chief digital officer and considering the job done, as many of the executives in that closed-door session were inclined to do, is the opposite of that. It’s a way of engaging with the transformation without actually doing it.

5. Raise the talent threshold deliberately and continuously.
When AI absorbs the execution work that once defined junior and middle layers, the people who remain need to be operating at a genuinely higher level. At Kiwi, there are no junior engineers left. The bar is being raised systematically — as a structural recognition that the work now demands it.

That last point is where this whole transition lives or dies. In my work with management teams across sectors and geographies, and in our forthcoming book A-Players: How to Hire the People Who Matter Most in the New World of Work, written with Jessica Neal and Ram Charan, we have found that the leaders who thrive in this environment share one quality above all others: they reframe the problem before the data forces everyone else to. That capability was always valuable. It is now decisive, because the layers that once absorbed the consequences of unclear thinking are gone.

Exhibit 2: Who entered the top 25 — and who left.⁹

Each card is a mega-cap firm in today’s top 50 by market cap. The left column ranks the top 25 by revenue per employee in 2015; the right column does the same for 2025. Curves connect firms that hold a top-25 slot in both years.

This is also where the question becomes interesting for investors. Most investors scrutinize talent at the top, and they would not put money into a company without it. Fewer look at organizational structure as an investment signal: whether the CEO is redesigning the company for what the technology now makes possible, or delegating the question to a function. That gap will not stay open for long.

The pyramid was the right answer to the right problem for a hundred years. But the problem it solved is going away. When organizing large workforces is no longer the central challenge of business, “organizations” — as we have come to know them — may no longer be the right word for what we build. What comes next is something closer to a living system: small, adaptive teams coordinating intelligence rather than managing people.

This shift will profoundly change how people live and work. More people will spend their working lives creating value rather than managing and administering, but millions of entry-level and middle-layer jobs will be displaced in the process. The human cost will be significant, and as a society, we will need to think seriously about how we support people through it.

For companies, there will be no choice but to adapt: those that don’t will be competed away by those that do. But the ones who see this change clearly, and build accordingly, will be very difficult to catch.

Methodology
Mega-cap revenue and employee counts from company 10-K filings via SEC EDGAR. Cohort for the flow diagram is the top 50 firms in the S&P 500 by market cap as of April 2026 (yfinance snapshot), held fixed across years. Revenue per employee = annual revenue / period-end full-time-equivalent employees as disclosed. A handful of cohort firms have missing employee counts in the 2015–2017 panel and are simply omitted from those years’ rankings (cohort coverage: 47–50 firms per year). Private-company benchmarks (Cursor, Anthropic, OpenAI, Lovable) are most-recent reported ARR over disclosed headcount, sourced from Sacra, TechCrunch, Bloomberg, and company announcements; these are run-rates, not audited revenue. Compiled 2026-04-30.

References
¹ Nevins, A. (1954). Appendix III. In Ford: The Times, The Man, The Company (pp. 645-648). Charles Scribner’s Sons. https://doi.org/10.1017/S0003055400272407
² General Motors Corporation. (1960). Statistical Summary. In GM Annual Report 1959 (pp. 42-43). General Motors Corporation. https://hdl.handle.net/2027/osu.32435027473867
³ S&P Compustat. (2026). IBM 1985 Revenue and Payroll. S&P Global Market Intelligence. Retrieved April 28, 2026, from https://www.marketplace.spglobal.com/en/datasets/compustat-financials-(8)
⁴ Amazon.com. (2009). Part 1. In 2008 Amazon.com Annual Report (4, 30). Amazon.com. https://s2.q4cdn.com/299287126/files/doc_financials/annual/Amazon_Annual_Report_2008.pdf
⁵ Nvidia Corporation. (2026). Part 1. In 2026 Nvidia Form 10-K (11, 37). Nvidia Corporation. https://s201.q4cdn.com/141608511/files/doc_financials/2026/q4/10K-NVDA.pdf
⁶ Nolan, B. (2025, May 9). Mark Zuckerberg runs Meta by only managing a ‘small group’ of 30. Fortune. https://fortune.com/2025/05/09/mark-zuckerberg-meta-managing-small-group-no-one-to-ones/
⁷ Fridman, L. (2026, March 23). Jensen Huang: NVIDIA – The $4 Trillion Company & the AI Revolution | Lex Fridman Podcast #494. YouTube. Retrieved April 29, 2026, from https://youtu.be/vif8NQcjVf0?t=337
⁸ Mega-cap revenue and employee counts from company 10-K filings via SEC EDGAR. Private-company datapoints reported by Sacra, TechCrunch, and Bloomberg, and company announcements (Anthropic Series G filing, March 2026; Lovable headcount disclosure, March 2026). ARR figures are run-rates, not audited revenue. NVIDIA’s fiscal year ends January, so its FY2025 covers a calendar period roughly 9–12 months earlier than the other mega-caps’ FY2025. Compiled 2026-04-30.
⁹ The cohort is held fixed at today’s top-50 by market cap, so the question is not “who joined the mega-cap club” but “who became more people-efficient than their peers.” The newcomers tilt heavily to firms that have absorbed AI as a factor of production — semiconductor and networking firms supplying the AI buildout, cloud-native software, capital-light platforms. The exits are firms whose revenue still depends on layers of human intermediation: payment-network back offices, retail operations, traditional financial services.

The views expressed in this piece are solely those of the author and are provided for informational purposes only. They should not be construed as investment, financial, legal, or other professional advice. This is an opinion article and does not constitute a recommendation or endorsement of any particular strategy, security, or investment. Readers should conduct their own research and consult with qualified professionals before making any financial decisions. GA does not accept or assume responsibility to you or any other person in connection with the provision of the Information. You acknowledge that you have made, and will continue to make, your own investment decisions without reliance on any representation or warranty of, or advice from, GA or its representatives. Kiwi is a portfolio company of General Atlantic. Kiwi is not a representative sample of GA’s current and former investments. A list of all current and former investments from 2011–present can be found on our website at www.generalatlantic.com.