Coordination Is the Bottleneck

Three weeks ago, we wrote that Block cutting 4,000 employees wasn't an AI story. It was a coordination cost story. The speed gain came from removing the communication overhead that thousands of people generate just by being part of the system. Brooks figured this out in 1975. Block rediscovered it in February 2026.

On March 31, Jack Dorsey and Roelof Botha published "From Hierarchy to Intelligence" — a 5,000-word organizational blueprint that says the quiet part out loud. The February layoffs weren't a cost cut dressed up with AI language. They were the first move in dismantling the management hierarchy entirely and replacing it with machine-driven coordination.

That framing matters. Because the narrative the market ran with in February — "AI makes companies leaner" — was wrong in a specific, important way. Leaner wasn't the point. Faster information flow was the point. And the thing blocking information flow was never the code. It was the org chart.

Why do companies default to hierarchy?

Dorsey's post traces hierarchical organization back to the Roman Army. Eight soldiers in a contubernium, eighty in a century, five thousand in a legion. Every layer existed to solve one problem: a human leader can effectively coordinate three to eight people. Stack enough of those layers and you can run an army, a railroad, or a 10,000-person fintech company.

The Prussian military formalized it after Napoleon embarrassed them in 1806. They created a dedicated class of officers whose job wasn't to fight — it was to plan, process information, and coordinate across units. Middle management before anyone called it that. American railroads adopted the same structure in the 1850s. So did every corporation that followed.

Every modern attempt to escape hierarchy — Zappos with Holacracy, Valve's flat structure, Spotify's squads — eventually reverted. Not because hierarchy is good. Because no alternative could route information at scale without it.

Dorsey's argument is that AI changes that equation. Not by making workers faster, but by replacing the routing function itself. A continuously updated model of the entire business — what's being built, what's blocked, where resources sit, what's working — eliminates the need for humans to relay that context through layers of management.

Does AI replace engineers or coordination?

This is the part that February's coverage got wrong, and it's the part that matters for every company watching Block's moves.

The standard narrative: AI tools make engineers more productive, so you need fewer engineers. That's the "copilot" story. Give everyone an AI assistant, same structure works with fewer people. Incremental improvement. Fine.

Dorsey's post explicitly rejects that framing. "Most companies using AI today are giving everyone a copilot, which makes the existing structure work slightly better without changing it. We're after something different."

The different thing: AI doesn't assist workers within the hierarchy. AI replaces the hierarchy's coordination function so workers don't need it. Individual contributors get context directly from the system instead of waiting for it to travel up and down a chain of command. No status meetings to prepare for other status meetings. No alignment sessions. No sprint ceremonies that exist to synchronize people who shouldn't need synchronizing.

Block's proposed structure has three roles. Individual contributors, directly responsible individuals who own cross-cutting problems for fixed periods, and player-coaches who build things while developing people. No permanent middle management layer. The system handles alignment. The DRI structure handles priority. Everyone else builds.

That's not "AI makes coding faster." That's "AI makes the organizational layer that coordinates coding unnecessary."

What do faster iterations actually look like?

Zenpo is a small company, not a 6,000-person fintech. But the dynamic Dorsey describes isn't theoretical — it's observable at every scale.

Zenpo is building a clinical trials platform right now. Not a simple one. The kind of system where business model, data architecture, platform architecture, and competitive positioning all have to evolve together. In a traditional setup, those workstreams gate each other. You lock the business model, then design the data architecture to support it, then build the platform on top. Sequential. Each transition requires alignment meetings, review cycles, stakeholder sign-offs before the next layer can move.

That's not what's happening. Business feasibility analysis is at its fourth iteration. Data architecture is on version three. Moat analysis is at version two. Platform architecture is evolving alongside all of them. These workstreams are advancing in parallel, at different speeds, co-evolving — and nobody scheduled a meeting to make that happen.

The version counts are the point. A business model at v4 means the assumptions have been challenged, stress-tested, reworked, and refined four times. That used to take months of scheduled reviews and alignment sessions just to get through the cycle of "here's the model, here's the feedback, here's the revision, here's the re-review." Now those iterations happen fast enough that the data architecture can respond to business model changes in near real-time instead of waiting for the quarterly strategy deck to finalize.

The bottleneck was never writing the code or drafting the analysis. The bottleneck was the coordination between layers — the meetings to align business and technical teams, the waiting for one workstream to stabilize before the next one could start, the overhead of keeping six interdependent tracks synchronized through human ritual. When that coordination compresses, the iteration velocity across the entire system accelerates in ways that have nothing to do with lines-of-code-per-hour.

That's the feedback loop collapsing in a productive direction. Not the dangerous collapse we wrote about in March — where market incentives reward headcount cuts regardless of operational health. The useful collapse, where information that used to require a standing meeting now travels through shared context.

This is why Zenpo has been biased toward lean teams since its inception in 2007, even for enterprise-level complex applications. Fewer people means fewer coordination points. Fewer coordination points means fewer miscommunications. Fewer miscommunications means fewer corrections. The clinical trials platform isn't lean-staffed because of budget constraints — it's lean-staffed because small senior teams with shared context outdeliver large teams drowning in alignment overhead.

Does AI-driven org design hold up long term?

Nobody knows if this works long-term.

Block is running the experiment at scale, and Dorsey's post is honest about what's aspirational versus what's operational. The "intelligence layer" that proactively composes financial products for individual customers based on transaction patterns — that's a vision, not a shipped feature. The world model that replaces management's information-routing function — that's a direction, not a benchmark.

Zenpo's own experience confirms the short-term gains. Fewer coordination rituals, faster iteration cycles, more time spent on the work instead of on talking about the work. Whether that compounds or hits a wall at some team size or complexity threshold is an open question.

The failure mode we flagged three weeks ago still applies. If companies use this framing to justify cuts beyond what the organization can absorb — if "AI replaces coordination" becomes the new "AI replaces workers" as a blanket justification — the same feedback loop collapse plays out. Stock goes up, operational cracks appear six months later, the narrative has already been set.

What did Block's layoffs actually change?

The February layoffs told you Block was cutting people. The March blueprint tells you why. Not because AI writes code faster than humans. Because the organizational layer that existed to coordinate humans — the meetings, the managers, the alignment chains, the status reports — turns out to be the actual bottleneck. And for the first time, there's a technology that can absorb that function.

The companies that will get this right aren't the ones announcing layoffs and citing AI. They're the ones quietly noticing that their teams move faster when coordination happens through shared context instead of scheduled rituals — and reorganizing around that observation instead of around a stock price.

Dorsey published the theory. The test is whether the practice survives contact with the reality of running a company at scale without the structure that every organization in history has fallen back on.