Organize Around Intelligence, Not Hierarchy

The org chart is a coordination technology. It exists because someone, somewhere, decided that the cheapest way to move a decision from the front line to the people accountable for it was to stack humans in a pyramid and pass information up the layers. For roughly a century that was true. It is becoming less true by the quarter, and the firms that internalize this first will be built on a different primitive entirely: not a chain of command, but a fabric of intelligence — human and agentic — wired around the work itself.

That reframing is the seed this piece grew from. It surfaced in The New Era of Jobs: Organizational Singularity with Peter Diamandis and Salim Ismail, where Ismail put it plainly: the AI-native company “needs to be architected around intelligence, not around hierarchy.” It’s a clean line. The interesting question for builders is whether the evidence supports treating it as an engineering instruction rather than a slogan — and where it breaks.

The mechanism: hierarchy is mostly coordination, and coordination is what’s collapsing #

Start with why the pyramid exists in the first place. Strip away the status and the org-design folklore, and a management hierarchy is largely a machine for coordination: aggregating reports, routing decisions, reconciling priorities across teams. A 2026 arXiv paper by Alex Farach frames AI precisely this way — as “coordination-compressing capital,” technology whose primary effect is not automating individual tasks but collapsing the cost of managerial coordination itself (arXiv). That matters because coordination is the load the hierarchy was built to bear. Drop the cost of the load, and the structure built to carry it becomes overhead.

You can see what replaces it in how the most aggressive adopters describe themselves. McKinsey has started using the term “work charts” — networks of human-plus-agent teams organized around exchanging tasks and outcomes rather than reporting lines — and notes that some early firms now express workforce composition in both FTE headcount and number of deployed agents (McKinsey). (McKinsey’s site is bot-walled, so I’m relying on the article as confirmed through multiple independent secondary summaries rather than a direct read; the “work charts” framing and the FTE-plus-agents accounting both check out across them.) The unit of organization stops being “who reports to whom” and becomes “which intelligence — wet or silicon — is best placed to do this and be accountable for it.”

This is not a fringe read. A 2025 MIT Sloan Management Review study run with BCG found that 69% of surveyed experts agree agentic AI demands fundamentally new management approaches, not mere adaptation of existing frameworks — because agents “operate independently, are goal-oriented, and have memory and reasoning capabilities, making their decisions complex, autonomous, and opaque” (MIT Sloan). You don’t manage that workforce with a reporting chain designed for humans who go home at 6pm.

The role that didn’t exist in a pyramid #

The cleanest signal that something structural is happening is the appearance of a job title that makes no sense in a hierarchy-first firm: the agent manager. A February 2026 Harvard Business Review piece documents the role directly — a person whose work is continuous dashboard oversight of an AI agent fleet, not management of people. At Salesforce, one such manager already oversees agents across support, sales, and marketing, starting and ending his day, in his own words, “in dashboards, scorecards, and agent observability monitoring” (HBR). That is a node in a work chart, not a box on an org chart. Its existence is the thesis made concrete.

The macro picture, stated honestly #

It’s tempting to reach for a big punchy adoption stat here. I’m deliberately not going to, because the most quotable one floating around — that “82% of business leaders call this a pivotal moment” — does not survive scrutiny. I couldn’t trace it to a primary source I trust, so I’m leaving it out rather than launder it into this piece. The pattern that is well-attested is more sobering than any single headline number anyway: deployment has gone nearly universal — agents are being bolted onto something in almost every large org — while genuine, measurable bottom-line impact remains the exception, not the rule. That gap — near-total deployment, scarce real impact — is the actual story. It’s exactly what you’d expect if firms are bolting agents onto hierarchies instead of rearchitecting around them.

A vendor-commissioned study points the same way, and it’s worth flagging the conflict of interest up front: MIT Technology Review Insights, in sponsored partnership with the AI vendor Ema (which sells an “Agentic Business Transformation” framework), reports that 85% of organizations want to be agentic within three years while 76% concede their current operations can’t support that shift (MIT Technology Review, sponsored content). Treat those numbers as directional marketing, not independent research — but the direction is the same structural mismatch McKinsey’s independent figures describe.

The concrete case: Amazon takes a hierarchy apart on purpose #

The clearest large-scale, publicly documented move is Amazon’s. In September 2024, Andy Jassy named the problem in the thesis’s exact terms: management layers had become organizational drag — preparatory meetings, review chains, deferred decisions. His response was to mandate at least a 15% increase in the individual-contributor-to-manager ratio by the end of Q1 2025, explicitly to push decision authority to the front line and strip out bureaucratic checkpoints (Amazon). (The memo sets the target; it doesn’t report the outcome, and I’ve seen no cited source confirming the goal was hit — so read this as a mandate, not a result.) For a company north of 1.5 million people, even setting that target is one of the largest deliberate de-layering experiments in corporate history.

Shopify institutionalized the same instinct at the point of resource allocation. Tobi Lütke’s April 2025 company-wide memo required teams to prove AI cannot do a job before requesting additional human headcount — making intelligence the default unit of work and people the exception you justify (CNBC). (CNBC bot-blocks direct fetches; the claim is confirmed via the outlet’s own search snippet, so the link may look broken if you click through.)

The honest counter-section #

The thesis is directionally right and overstated in the usual ways. Four corrections worth holding:

Coordination is only one of hierarchy’s jobs. A 2023 paper in the Journal of Organization Design argues that even if AI zeroes out coordination cost, firms still need hierarchy to assign accountability, resolve disputes, and signal authority to outside stakeholders (Springer). The structure doesn’t vanish; it shrinks and shifts upward. “Architected around intelligence” still needs a human whose name is on the outcome.

Flat is not the same as intelligently organized. The same literature points to Valve’s famously bossless structure producing creative inertia and stalled projects, because informal power fills the vacuum left by absent formal authority. Removing the hierarchy is the easy half; replacing its coordinating function with something better is the hard half — and skipping the hard half doesn’t get you an AI-native firm, it gets you a leaderless one.

Compression can entrench hierarchy, not flatten it. Farach’s paper has a “regime fork”: under different ownership conditions, the very same coordination-compressing capital concentrates authority in superstar managers overseeing ever-larger agent fleets. In simulation, the manager-worker wage gap widens from about 1.8× to 5.4× at high agent-capital adoption (arXiv). Organize around intelligence and you might get a pancake — or a much steeper pyramid with fewer, more powerful people at the top. Who owns the elasticity decides which.

Wholesale replacement breaks where judgment lives. Klarna cut roughly 700 customer-service roles in 2024 chasing ~$40M in savings, then reversed course as satisfaction dropped on complex cases, repeat-contact rates rose, and lost institutional knowledge proved costly to rebuild. It settled on a hybrid — AI handling 60–70% of tier-one volume, humans owning escalations (Digital Applied). And the pipeline problem is real: Yale’s Tristan Botelho cautions, “I don’t think middle management is going to be erased. I think it’s going to just redefine how managers think about their role” (Fortune).

What to build with this #

For operators, the instruction survives the caveats: stop optimizing the reporting chain and start designing the intelligence fabric. Map your work charts before your org charts — for each workflow, ask which intelligence is best placed to run it and which human owns the result. Make AI the default unit of work and headcount the exception you justify, Shopify-style. Build the agent-manager function deliberately rather than letting agents proliferate ungoverned into Valve-grade entropy. And keep the accountability shell intact: the firm is still a liability sink and a name-on-the-outcome container, and no amount of cheap intelligence dissolves that.

Hierarchy isn’t dying because pyramids are bad. It’s thinning because the thing it was built to carry — coordination — is getting cheap fast, and the firms that win will rebuild around what stays scarce: judgment, accountability, and the ability to learn faster than anyone else.