Something happened this week that I want to think through with you.
Jack Dorsey and Roelof Botha — the heads of Block, the company behind Square and Cash App — published an essay called From Hierarchy to Intelligence. It is not a typical corporate white paper. It articulates a theory of what companies are — and then proposes that the answer is about to change.
I've been working on a book about this same territory since the end of 2023. It's called Undesigned: How Agentic AI Is Rewriting the Logic of Organizations, and it's currently in its third draft. Reading Block's essay felt like watching someone arrive at a familiar destination from the opposite side of the mountain. Same view, a completely different path. The landscape each of us can see from our respective routes don't entirely overlap. That's what makes the conversation worth having.
Block's argument is elegant and historically grounded. They trace the organizational hierarchy from its origins — the Roman contubernium, the principle that a leader can effectively manage three to eight people — through the Prussian General Staff, the American railroad, Frederick Taylor's scientific management, the Manhattan Project, McKinsey's matrix organization, and the recent experiments at Spotify, Zappos, and Valve. Two thousand years of organizational innovation, they argue, has been a single sustained attempt to work around a fundamental constraint: narrowing your span of control requires adding management layers, but more layers slow information flow.
Every organizational form in history has been a different answer to the same question: how do you coordinate collective effort without drowning in your own coordination overhead?
Block's answer is sort of contrarian. Rather than augmenting existing hierarchies with AI copilots (the approach most enterprises are taking), they want to build what they call "a company organized as an intelligence." The idea rests on two pillars. A company world model, built from the digital exhaust of remote-first operations where every decision, discussion, and design exists as a recorded action, and a customer world model, built from Block's unique position on both sides of millions of financial transactions daily.
The structural implications follow naturally. Three roles replace the traditional hierarchy: Individual Contributors (deep specialists), Directly Responsible Individuals (who own cross-cutting problems with authority to pull resources across teams), and Player-Coaches (who combine building with people development). The explicit conclusion: "There is no need for a permanent middle management layer."
Intelligence moves to the system. People move to the edge, where humans can handle what models cannot: intuition, ethical reasoning, cultural context, the feeling in a room. It's a vision worth taking seriously, I believe.
In Undesigned, I approach this same underlying idea from a different direction. Rather than starting with the org chart and asking what AI replaces, I start with the AI capabilities themselves and ask what happens to organizations as those capabilities mature. The path is different, but we're drinking from the same well. And there are places where the view from my side of the mountain adds something to the picture — not because it's better, but because it catches light that the other angle doesn't.
One of those places is what I call the familiarity trap.
Old lenses
Block's essay reads as if the primary obstacle to this transformation is conceptual. Once you see the organizational chart as an information-routing protocol and recognize that AI can do the routing better, the path forward becomes clear. Hard, certainly, but essentially an engineering and design challenge.
I think there's something else going on. The familiarity trap and it is not exclusively about executives who don't understand AI. I see a cognitive pattern that produces perfectly rational outcomes at every individual decision point while quietly generating a blind spot at the level of strategic positioning.
This pattern works through three reinforcing forces. The first is the pressure to demonstrate measurable returns. Task automation delivers ROI beautifully — you can count the hours saved, the throughput increased. Structural transformation is maddeningly difficult to quantify in advance. How do you build a business case for reorganizing your entire company around AI-driven world models when you cannot predict what those models will discover? The ROI framework, as I write in the book, "rewards exploitation of known processes. It penalizes exploration of unknown structures."
The second force is opacity. Foundation models are not fully understood by the people who build them, let alone the executives who deploy them. Again: this is not ignorance about the capability of these systems, but a biased inclination towards bounded applications — the ones you can benchmark and monitor. The agentic frontier, where systems coordinate autonomously in open-ended environments, is precisely where opacity is most concerning. So organizations retreat to territory they can control. Reasonable risk management becomes, in aggregate, a systematic avoidance of structural possibility.
The third force is cognitive framing — and it's the deepest. We interpret new phenomena through existing mental models. When a CEO sees an AI agent resolving support tickets, the existing frame (task, worker, output, efficiency) maps perfectly. When that same CEO hears about an AI system that generated its own research hypothesis, coordinated sub-agents to test it, and revised its approach without any human instruction, then the existing model strains. The instinct is to either dismiss it or force it into the familiar frame: "So it's like a very autonomous employee." But it is not like a very autonomous employee. It is like something we do not yet have a category for.
These three forces reinforce each other. ROI pressure directs investment toward automation use cases. Manageable opacity confirms the existing mental model. The confirmed mental model shapes the next round of ROI analysis. As I describe it in the book, the cycle is "stable, productive, and entirely capable of sustaining itself while the structural transformation of entire industries happens somewhere outside the frame."
I think Dorsey and Botha have clearly seen past this trap — you simply can't publish an essay like theirs from inside it. But much of their audience hasn't. My honest suspicion is that many executives will read From Hierarchy to Intelligence, nod along, admire the vision, and return to optimizing their existing operations with copilots. Not because they're wrong in any given quarter. But because:
The executive who chooses the safe AI deployment is not wrong in any given quarter. The danger is that the quarters accumulate, and by the time the structural shift becomes undeniable, the organizational capacity to respond to it has atrophied. The trap is not in any single decision. It is in the pattern.
From Undesigned, Chapter 1 (unpublished draft).
Who do you think you're talking to?
There's another dimension here that I've been sitting with for a long time, and that Block's essay brought back into sharp focus.
When they write that in their new model "the intelligence lives in the system" and "the people are on the edge," it sounds like a statement about information architecture. But there's something underneath it that goes deeper — something about power.
Organizations are, at their core, systems for distributing epistemic authority. Who knows what. Who is credentialed to decide what, and whose judgment is trusted, and under what circumstances. The entire apparatus of hierarchy — titles, reporting structures, committees, approval chains — is fundamentally an architecture for managing who has the right to claim knowledge and act on it.
This architecture was designed for a world in which knowledge was scarce, hard-won, and embodied in human beings. Senior people knew more because they had seen more. Expertise was a function of experience, and experience was a function of time. A thirty-year veteran carried knowledge that could not be extracted or transferred to a junior colleague over a weekend.
What happens to this architecture when a system with no real-world experience at all can, in many domains, outperform a human with thirty years of sweat and tears?
This is not hypothetical. A 2024 study affiliated with Harvard Business School tested AI systems against human CEOs on strategic decisions. The AI outperformed the humans on most dimensions — more consistent, less susceptible to cognitive biases, better at integrating large quantities of information. The humans performed better only on genuine black swan events, where no historical data applied. In Undesigned, I try to sit with what this actually means:
The organizational concept of expertise itself is undergoing fundamental redefinition. Expertise has meant authority to make decisions, credibility to challenge courses of action, and standing to mentor colleagues. Every function depends on superior knowledge — now challenged.
From Undesigned, Chapter 3 (unpublished draft)
When Block says "there is no need for a permanent middle management layer," that's an architectural claim. But the human experience of it is not architectural, it is existential. In the book, I look at cases like Capital One, where fraud analysts resisted agent-generated alerts because accepting them meant accepting that the expertise which defined their place in the organization was no longer the scarce resource it had been. Or Dow Chemical, where senior scientists found themselves unable to evaluate the flood of AI-generated hypotheses — not a failure of adaptation, but the disorienting experience of watching your professional role transform from generating insight to curating insight generated by something else.
Block's three-role model (ICs, DRIs, Player-Coaches) is a thoughtful design for routing authority in a post-hierarchical organization. But it carries the implicit assumption that the people filling those roles will have made peace with a reality that most professionals have not yet confronted: that the knowledge which once defined their value is no longer uniquely theirs.
I don't say this as a critique. I think Block is asking the right questions and proposing interesting answers. But the transition from here to there runs through territory that organizational design alone can't navigate. It requires something closer to institutional honesty, a kind of acknowledgment that this shift is not merely structural but deeply personal. The organizations that navigate it well will be the ones that give people time and space to build a professional identity around capabilities that machines genuinely cannot replicate. Block, to their credit, puts these capacities at the edge. I am just acknowledging that getting there will be a slow and messy process for all humans involved. And that it is not certain how the current anchors of authority and power, specially experience, will play out in this new world.
And here is where the conversation extends beyond Block and corporations.
The World as organizational infrastructure
If AI can build world models of businesses, it can build world models of cities. If it can replace the information-routing function of middle management, it can replace the information-routing function of bureaucracies. If self-organizing agent systems can coordinate without centralized control, the question "what is the function of an organizational chart?" becomes "what is the function of an institutional hierarchy?"
Including the ones we call government.
This is not distant futurism. The capabilities are installed. Different AI agents, running on different infrastructure, can currently discover one another, exchange context, and coordinate action.
If agents coordinate across platforms easily, what defines firm boundaries? Ronald Coase argued firms exist because market transaction costs exceed internal hierarchy costs. Agent protocols reduce coordination costs across boundaries dramatically.
From Undesigned, Chapter 4 (unpublished draft)
If the Coasean boundary of the firm dissolves, the same logic applies to any institution that exists because coordination is expensive. Municipal governments coordinate public services across departments through bureaucratic hierarchy because that was the only available coordination technology for systems of that complexity. National level, all the same. International governance structures exist, in part, because coordinating across jurisdictions requires formalized protocols that are slow by design.
None of these institutions was designed for a world where coordination can be automated, where world models can maintain continuous awareness of complex systems, where agents can negotiate and adapt in real time. The question is not whether AI will eventually touch governance. It's whether the institutional imagination exists to use it for something more than efficiency gains within existing structures.
Imagine a city where an AI world model — not unlike what Block describes for its business — maintains continuous awareness of infrastructure, services, resource flows, and citizen needs. The upside of institutional surveillance, in which information-routing benefits, serves people.
At least technically, this is not utopia. It carries its own risks, but the possibility is worth holding alongside the risk, because the alternative — maintaining institutional structures designed for the information constraints of the nineteenth century — carries risks of its own. The gap between what institutions could do and what they are doing is widening. And that gap has consequences, measured in services undelivered, patterns undetected, and citizens whose needs fall through the cracks of a coordination architecture that was never designed for this level of complexity.
Block is asking a very elegant question for corporations: what does your company understand that is genuinely hard to understand, and is that understanding getting deeper every day? The equivalent question for public institutions might be: what does your institution need to understand to serve its constituents, and are your current structures capable of understanding it?
If the answer is no, then AI doesn't merely offer optimization. It offers a different kind of institutional intelligence, designed with care and governed with wisdom, we can nurture dreams of more meaningful participation, more responsive services, and more informed collective decision-making than any hierarchy of human administrators can achieve alone. Hello again, democracy.
This was, by the way, something we talked about in the 1990's, when the internet was reaching critical mass.
An idea getting shape
I want to be clear that I'm not saying this will or should happen. What I am saying is that the convergence around this particular idea of AI-enabled (re)organizations is worth noticing. When a major corporation publishes a thesis about replacing hierarchy with intelligence, and an independent researcher and writer working alone from the opposite direction arrives at structurally similar conclusions, that coincidence is itself a signal. The kind of signal futures thinkers love to pay attention to.
Most likely, the future of organizations is not going to be designed by any single architect. It's going to emerge. Many of these experiments will fail. The ones that work will reshape not just how companies operate, but how institutions of every kind coordinate collective human effort.
The familiarity trap is real. The human dimension of this transition already looks and feel harsh. The underlying capabilities — reasoning, self-organization, simulation, replication, self-scaffolding — are not waiting for institutions to be ready.
The question I'll insist asking is whether we'll meet them with the imagination they demand.
Undesigned: How Agentic AI Is Rewriting the Logic of Organizations is my new book, work in progress. Excerpts shared here are from the unpublished third draft (March 2026).