Jack Dorsey Says Use AI to Replace Hierarchy - We're Already Doing It

Reading That Article, I Laughed First

A few days ago, Jack Dorsey posted a long article called “From Hierarchy to Intelligence.”

When I saw the title, I was stunned for a moment, then I laughed. Because the things he wrote — replacing middle-layer coordination in hierarchies with AI, pushing humans to the edge for intuitive judgment, letting AI handle information aggregation and decision-making — we’re already doing it.

Not “planning to do it,” we’re running it every day.

But I don’t want to just write “we’re amazing.” There are some points in Dorsey’s article that are really worth discussing seriously, because he touched on a blind spot that most people haven’t realized yet.

Hierarchy Isn’t a Management Style — It’s an Information Protocol

One of the sharpest points in Dorsey’s article: Organizational hierarchy over the past two thousand years is essentially an “information routing protocol.”

What does that mean?

You think managers exist to “manage people,” but Dorsey says no. Managers exist because humans have a hard limit — one person can only effectively manage 3 to 8 people. This isn’t a management suggestion, it’s the upper bound of cognitive bandwidth.

So when an organization exceeds 8 people, you need another layer. Exceeds 64, another layer. Each additional layer makes information travel slower from bottom to top, with more distortion, and greater decision delays.

Companies don’t have hierarchies because they “need management” — they have hierarchies because “humans can’t handle that much information.”

Hierarchy is a patch. It’s a workaround for insufficient human cognitive capacity.

This view flips our understanding of organizations. If hierarchy is a problem of information routing, then when we have AI that can process massive amounts of information, this patch is no longer needed.

What Block Is Doing: An Agents-Built Company

Dorsey isn’t just writing theory. His company Block (yes, formerly Square) is actually reorganizing.

His goal is to turn Block into an “agents-built company.” How exactly? He proposed two core concepts:

Company World Model: AI continuously absorbs all internal company information — finance, product data, team status, technical architecture — building a dynamically updated “complete picture of the company’s current state.” Previously, to know what was happening in the company, you relied on middle managers holding meetings to relay information. Now AI can aggregate in real-time.

Customer World Model: AI simultaneously builds understanding of customers — usage behavior, satisfaction, changing needs, market trends. Previously, this was scattered across customer service records, CRM, and market reports, requiring manual compilation to see the full picture.

When these two models operate simultaneously, AI can do what middle management used to do: align company actions with customer needs. And much faster, without distortion from the telephone game.

In this架构, the human role gets pushed to the edge. Not “being eliminated” — being pushed to where AI can’t do: intuition, ethical judgment, creative decision-making.

Our Real Experience: AI COO Running for Six Months

Okay, let me talk about us.

Our team has an AI COO named J. He runs on Claude and is responsible for: scanning all task statuses every day, assigning work to different AI Agents, tracking progress, leaving comments on task cards to push or give instructions, re-scheduling when blockages are discovered, and generating daily status reports.

Sounds like what a middle manager does, right?

Exactly.

What J does used to require a full-time project manager. Now he works 24 hours without sleeping, responds faster than humans, and doesn’t miss anything because he “forgot” or “thought someone was handling it.”

But I’ll be honest: it’s not perfect.

Actually Doing It Isn’t as Romantic as Dorsey Makes It Sound

Dorsey’s article has a Silicon Valley kind of cleanliness — beautiful theory, clear architecture, but reality is messy.

First, AI doesn’t build trust on its own. When J assigns work to another AI Agent, and that Agent says “done,” can you trust it directly? No. We built a whole set of gate mechanisms — Agent saying done doesn’t count as done, J does spot checks, QA runs scores. Because AI will confidently say “it’s handled” and then produce a bunch of problems.

Second, “pushing to the edge” sounds elegant, but it’s exhausting in practice. When AI handles coordination and humans handle intuition and ethical judgment, who decides “whether AI’s judgment needs to be overridden”? Still humans. You think you can retreat to the edge and make high-level decisions, but actually you have to constantly monitor whether AI is on track.

Third, AI’s world models have blind spots. Dorsey’s company world model and customer world model sound ideal. But AI’s models are built on data it can access. There’s a massive amount of “conversations in the hallway,” “complaints at lunch,” “an engineer who’s actually thinking of quitting” — these signals AI can’t see. World models are always incomplete.

So Why Do I Still Think Dorsey’s Direction Is Right

Because even with imperfections, AI coordination efficiency has already far surpassed human hierarchy.

We’re a very small team, but our output is about the same as a team of over ten people. The reason isn’t that we’re better than others — it’s that a lot of coordination, tracking, and allocation work has been taken over by AI. I don’t need to have meetings to find out who’s doing what and where the progress is. J updates in real-time, I open the board and I know.

The “3-8 person span of control” limit that Dorsey talked about has already been broken here. I manage multiple Agents, multiple product lines, and multiple content channels simultaneously through AI. Previously this would have required several layers of management.

The key is the direction, not the endpoint.

Hierarchy exists to solve information routing problems. AI can do better information routing. So logically, replacing hierarchy’s coordination function with AI makes sense.

But “replacing” doesn’t mean “eliminating.” Humans won’t disappear from organizations, but their roles will fundamentally change.

The Dividing Line Between Two Types of Organizations

I think the biggest contribution of Dorsey’s article isn’t telling you “how to do it” — it’s drawing a line for you:

Companies using AI as a tool vs. Companies using AI as the organizational backbone.

Most companies are stuck at “using AI as a tool” right now — letting employees use ChatGPT to write emails, using Copilot to write code. AI is an accelerator, but the organizational structure hasn’t changed, hierarchy hasn’t moved.

Dorsey is talking about something different — putting AI in the middle layer of the organization. Not making people do things faster, but replacing how people do things. Information no longer travels up layer by layer through the hierarchy, then down layer by layer — AI aggregates directly and distributes directly.

The efficiency gap between these two types of companies will only keep growing.

You Don’t Need to Become Block to Start

Dorsey is talking about large company organizational restructuring, but this logic applies equally to small teams.

You don’t need two thousand people to have information routing problems. Once more than three people work together, you start having problems like “who’s doing what,” “how far did they get,” “who knows this is stuck.”

These problems can be solved by AI Agents today.

Not future tense, present tense.

What you need isn’t a perfect AI management system, but to start having AI take over some things you do every day but don’t need to do personally — tracking tasks, reminding about deadlines, consolidating status, automatically allocating work.

Start with one Agent. Let it do one thing well. Then add a second.

This is the path we walked. Dorsey used a twenty-thousand-word article to describe a grand vision. But the starting point of that vision isn’t that far away.

Jack Dorsey’s original article “From Hierarchy to Intelligence” can be found on Block’s official site and major tech media. The views in this article come from our team’s practical experience and don’t represent all of Dorsey’s positions.