The Layer Dorsey Didn’t Name

From Span of Control to Network of Commitments

Brian Connelly

Jack Dorsey just told the world that hierarchy exists to route information, and AI routes it better. Five million people read that in two days. He’s not wrong about the routing. He’s wrong about what hierarchy actually does.

Here’s what everybody in this conversation is arguing about. Which architecture handles information best? Vector databases? Structured ontologies? Transaction signals? Those are real engineering questions and smart people are doing good work on them. But they’re all answering the same question: how do we move knowledge through an organization without humans carrying it in their heads?

That’s the wrong question. Or more precisely, it’s the second question. The first question is: what holds an organization together when the information is moving just fine?

The answer is commitments. Not data. Not dashboards. Not the world model. Commitments. When your VP of Product says we are shipping this in Q3, that is not information. It’s an act. It creates an obligation. Someone is now on the hook. And the entire apparatus of the organization, the plans that get made, the resources that get allocated, the other promises that get made downstream of that promise, all of it depends on that commitment being made, tracked, honored, or renegotiated when reality changes.

A world model can tell you what happened. It can even tell you what’s likely to happen. What it cannot do is make a promise. And it cannot be held accountable when the promise breaks.

Dorsey says money is the most honest signal in the world. I’d say something slightly different. Money is the receipt. The commitment already happened. Two people decided to trust each other enough to transact. The transaction is the exhaust. The relationship is the engine.

The Structure Nobody Sees

Now here’s where people get stuck. Everyone knows commitments matter. That’s not an insight, it’s a truism. The insight is that commitments aren’t random. They have a structure, and that structure is consistent enough to design around.

Think about what actually happens when work moves in an organization. Someone makes a request. Someone else makes a promise. Work gets done and delivered. The requesting party says yes, this is what I needed, or no, it isn’t, and the cycle starts again. Request, promise, deliver, accept. That’s it. That’s the atomic unit of all coordinated work. Not information. Not tasks. Conversations that end in commitments.

This isn’t new thinking. J.L. Austin identified it in 1955 when he distinguished between language that describes the world and language that changes it. John Searle formalized it further in 1969. When someone says “I’ll have that to you by Friday,” they haven’t transmitted information. They’ve performed an act. They’ve altered the social reality between two people. The entire field of speech act theory is built on this distinction, and it maps directly onto how organizations actually function.

Now scale that up. An organization of four thousand people isn’t four thousand individuals processing information. It’s thousands of these loops running simultaneously, nested inside each other, dependent on each other. Your VP promises the board a revenue number. That promise generates a cascade of downstream commitments across product, engineering, sales, and operations. Each of those commitments generates further commitments. The whole thing is a living network, and the network holds together only as long as the commitments in it are being made well, tracked honestly, and renegotiated when conditions change.

This is where the world model conversation gets interesting. Because a world model can see the artifacts of those commitments. It can see the documents, the Slack messages, the tickets, the revenue numbers. But it cannot see the commitments themselves. It doesn’t know the difference between a genuine promise and something someone said in a meeting to end the conversation. It doesn’t know that a commitment was made under pressure and the person who made it already knows they can’t deliver. It doesn’t know that a renegotiation needs to happen but nobody wants to be the one to surface it.

Those aren’t information problems. Those are commitment quality problems. And they are the things that actually determine whether an organization executes or stalls.

What Dorsey Got Right

Here’s what’s interesting. Dorsey actually gets closer to this than most of his critics realize. Block’s new structure has three roles: individual contributors who build, directly responsible individuals who own outcomes for a fixed period, and player-coaches who develop people and maintain culture. Most of the commentary has focused on what was removed. Four thousand people, the entire middle management layer, gone. But look at what was kept.

The DRI role is a commitment structure. A DRI who owns merchant churn reduction for ninety days is not an information processor. That person is making a promise: I will own this outcome. I accept accountability for it. Give me the authority and I will deliver a result. That’s a speech act. The world model didn’t generate it. A human being stood up and said I’ve got this.

And the player-coach role is commitment stewardship by another name. Someone who develops people and maintains culture is doing the work of making sure the humans in the system are capable of making and keeping good commitments. That’s not information routing. That’s tending the network.

So Dorsey kept the commitment layer. He just didn’t name it. And because he didn’t name it, he didn’t design for it. The world model tracks everything the DRI does. It does not track the quality of the commitment the DRI made, whether it was made well, whether it was made under the right conditions, whether it needs renegotiation, or whether the downstream commitments it generated are still coherent. Those questions fall into a gap between what the system sees and what the organization needs. And that gap is where decision quality will quietly degrade.

The Cost Nobody’s Counting

When this goes wrong, the ROI conversation will default back to transactions. That’s what bean counters do. And the dashboards will support the story because the world model is really good at counting things. Revenue, throughput, cycle time, tickets closed. All visible. All measurable. All green.

What the dashboards won’t show is the human carnage.

This is what has plagued every radical organizational transformation that tried to leap past the traditional models. Without strengthening the internal relationships that commitments depend on, stress and attrition skyrocket. The people who were holding the commitment network together, the ones who knew which promises were real and which were theater, who could walk into a room and renegotiate a deadline before it became a crisis, those people are gone. And the ones who remain are being asked to operate inside a system that tracks everything about their work except whether anyone has their back.

Even the military figured this out. The hardest, most operationally demanding organization on the planet learned centuries ago that the unit of survival is not the individual. It’s the bond between the person on your left and the person on your right. That bond is a commitment. I will not leave you. You can count on me. The chain of command exists to protect that bond, not the other way around.

Dorsey’s paper treats hierarchy as an information bottleneck that technology has now solved. The military learned the opposite lesson. Hierarchy exists to protect the commitment structure that holds people together under pressure. Remove the hierarchy without replacing that function and you don’t get a faster organization. You get an organization that looks fast on the dashboard and is quietly shredding the people inside it.

Remember the experiments of the 2010s. Zappos adopted holacracy and satisfaction scores collapsed. Valve’s flat structure hid a shadow power hierarchy that became a well-documented case study. Medium’s head of operations wrote publicly that the system was getting in the way of the work. Those failures were loud. Visible. Diagnosable.

The world model failure will be quiet. Attrition will tick up. The best people will leave first because they’re the ones who can. Institutional knowledge will walk out the door. And the system will backfill the data and keep running as if nothing happened because it doesn’t know what it lost. It only knows what it can count.

You can measure transactions all day. You cannot measure the commitment someone decided not to make because they no longer trust the system enough to put themselves on the hook. That’s the cost nobody’s accounting for. And no world model will see it coming.

Build the Model. Name the Layer.

None of this means the world model is a bad idea. The information routing problem is real and AI does solve it faster and cheaper than hierarchy. Build the world model. Automate the status syncs and the dependency tracking and the report generation. That work should have been automated years ago.

But when you remove a management layer, know what you’re removing. Map the commitment network before you restructure it. Know who is making promises to whom, which of those promises are load-bearing, and what happens to them when the people stewarding them are gone. Design the roles that remain, your DRIs, your player-coaches, around commitment quality, not just information access. And build the discipline of asking a question no dashboard will ever answer for you: are the people in this system still willing to put themselves on the hook for each other?

The world model sees everything your organization does. The commitment network is everything your organization is. You need both.