Five Gates Between Hierarchy and Intelligence

Last week Jack Dorsey and Roelof Botha published From Hierarchy to Intelligence, arguing that artificial intelligence can replace the 2,000-year-old coordination function that hierarchy has provided. They are correct.

The harder question is what to do about it.

Hierarchy Was Always Infrastructure

The central insight in the Block piece is easy to nod at and miss. Hierarchy is not an authority structure. It is an information routing protocol. The Roman decanus existed because eight soldiers in a tent needed someone to coordinate them. The Prussian General Staff existed because nineteenth-century armies could not move information fast enough without dedicated middle layers. Modern middle management exists for the same reason: human bandwidth limits required someone to aggregate context, route decisions, and translate strategy into execution.

Every prior attempt to flatten organizations failed at scale because no technology actually replaced what hierarchy was doing. Spotify squads, Holacracy, matrix structures: they all rearranged the chart without solving the coordination problem underneath.

AI is the first technology that genuinely can.

The Trap

A common pattern is already taking shape.

A senior leader reads the Block piece. They get excited. They announce a flatter structure. They eliminate a management layer. They deploy Claude or ChatGPT or Copilot to 'every employee.' They run a townhall about being AI-native.

Six months later, the meetings still exist. The status reports still get written. The decision queues still bottleneck on the same five executives. Productivity is marginally up because people are using AI as a faster typewriter. The intelligence layer that was supposed to replace coordination does not exist.

This is the trap. Block's vision is not about org charts. It is about replacing the operating model. If you flatten the structure without building the intelligence layer that makes flatness viable, you have not transformed anything. You have just removed the people who were holding the old system together.

Five Gates Between Hierarchy and Intelligence

The path is not a destination. It is five gates, each of which has to clear before the next one matters.

• Gate 1: Foundation. A real platform layer with SSO, RBAC, audit logging, observability, cost attribution. This is the boring infrastructure that determines whether AI gets used everywhere or stays in pockets. Without it, you cannot govern. Without governance, finance and security will shut you down within a quarter.
• Gate 2: Signal. Block calls it the company world model. The substance underneath is simpler: machine-readable artifacts of work. If your decisions live in meeting transcripts that nobody indexes, your priorities live in slide decks that nobody parses, and your operational state lives in dashboards that nobody connects, you have no signal. The intelligence layer needs something to be intelligent about.
• Gate 3: Composition. This is where MCP earns its keep. The hard part is not building one Claude integration. It is building dozens of tool-augmented workflows that compose existing systems into new capabilities without rewriting them. A sales workflow that pulls from CRM, email, and call transcripts to produce account intelligence is composition. A chatbot in your CRM is not.
• Gate 4: Edge. Block's most important sentence: 'A world model that can't touch the world is just a database.' The intelligence layer only matters if humans at the edge are using it to do real work and the work is feeding back into the system. This is an adoption problem, not a technology problem. If your team uses AI as a reference tool but executes the same way they did a year ago, the edge is dead.
• Gate 5: Trust. The hardest gate. Trust that the system has the right context. Trust that the model will not hallucinate into a contract or an outage. Trust that the audit trail is honest. Without trust, every output gets re-validated by a human, which means you have built an expensive layer of redundancy on top of your existing hierarchy. The org never actually flattens.

Each gate is a discipline. Companies that try to skip ahead build pilots that look impressive in a demo and collapse under production weight.

The Substrate Is Already Here

The unspoken assumption in the Block piece is that intelligence-as-infrastructure requires a multi-billion-dollar engineering investment. Block can afford that. The interesting question is whether anyone else needs to follow that path.

They do not.

Anthropic has shipped most of the substrate. Claude provides the reasoning core. The Model Context Protocol provides the integration fabric. Claude Code provides the agentic execution layer. Managed Agents provide the autonomous workflow runtime. The 1M context window means the entire company world model can fit in a single prompt for many use cases.

The work is no longer building the intelligence layer. It is composing it, governing it, and migrating into it. That is a meaningfully different problem than what Block solved, and it is solvable in months rather than years.

The companies that understand the difference will move first. The ones that read Dorsey and conclude they need to spend two years engineering their way to parity will spend those two years watching their competitors lap them.

We Are Not In Early Days

The framing matters. We are not in the early days of AI augmentation. We are in the late days of organizational hierarchy.

That distinction changes how you plan. If you think you are early, you wait. If you think the old model is ending, you move.

Block moved.

The question for everyone else is whether they understand the gates well enough to recognize what they are actually building.