← writing

The Org Is a Computer Built Out of the Wrong Hardware

April 14, 2026

Thesis

Every organization is a distributed computing system. It ingests information, processes it, routes it, stores it, and produces outputs. For all of human history, we've had no choice but to run this system on biological hardware — human brains. But human brains are slow, lossy, inconsistent, politically motivated, need 8 hours of downtime daily, and permanently delete their local state when they leave. We now have thinking machines that can run this same system faster, cheaper, losslessly, and 24/7. The org doesn't need to be redesigned. It needs to be re-platformed.

The Org as Network Architecture

Every role in an organization maps directly to a component in distributed systems:

Org Role Network Equivalent
IC (Individual Contributor) Edge node — where actual computation happens
Manager Router — receives packets from above, routes to correct downstream nodes
Director Load balancer / aggregator — distributes work, merges outputs
VP Cache layer — holds compressed, stale summaries of system state
CEO Root scheduler — sets global priorities, allocates resources
New hire Newly provisioned node
Onboarding State loading onto a fresh node
Someone quitting Node failure with total local state loss, no backup
Firing Decommissioning a node
Reorg Network topology change that breaks existing routing tables

The Four Operations

Everything an org does falls into four categories:

1. Context Onboarding (State Loading). A new human joins. They know nothing. The org spends weeks or months loading state into their brain — who does what, what matters, how things work, where things live. This is cp from a distributed filesystem into a new local node. It's slow, incomplete, and the new node is useless until it finishes.

2. Context Syncing (Synchronization Protocol). This is what meetings are. All of them. Monday standup: state sync. Quarterly business review: state sync. 1-on-1 with your manager: bidirectional state sync. Email: async state sync. Slack: real-time state sync. Humans spend 60-80% of their work week doing nothing but synchronizing state across nodes. The entire "collaboration" layer of work is just a very slow, very expensive sync protocol.

3. Machine Output (Computation). The actual artifacts. The code that gets written. The deck that gets built. The survey that gets sent. This is the REAL work — the thing the org exists to produce. It accounts for maybe 5-20% of total human time in the org. Everything else is overhead to make this possible.

4. Human Thinking and Direction (Scheduling / Priority Setting). Taste. Judgment. "What should we build and why." This is the only layer that's genuinely human. It sits above everything else. It's the thing that no amount of context routing can replace — the novel decision, the creative leap, the strategic bet.

The Query Model

Every Slack message, every email, every "hey, quick question" is a query against a human brain.

The entire communication layer of an org is humans querying other humans' brains and waiting for responses. Response times are measured in hours or days. Results are inconsistent. There's no schema. There's no cache invalidation. The "database" goes on vacation for two weeks and takes its data with it.

Failure Modes

Every classic distributed systems failure shows up in orgs:

System Failure Org Equivalent
Stale cache VP still operating on last quarter's strategy
Split brain Two teams building the same thing, neither knows
Single point of failure "Only Sarah knows how that works"
Byzantine fault The person actively lying in meetings
Network partition Two departments that stopped communicating
Data corruption Miscommunication that propagates through the chain
Race condition Two teams committing to conflicting plans simultaneously
Memory leak Meetings that grow in attendees but never shrink
Cascading failure Key person quits, three projects collapse
Packet loss "I never got that email"

The Compression Tax

"Make it higher level for the VP."

This is not a communication skill. This is lossy compression for a bandwidth-limited node.

The information hierarchy in an org is a compression pipeline:

Each level up is another lossy compression pass. By the time information reaches the top, it barely resembles reality. Then decisions based on that compressed signal flow BACK DOWN through the same lossy chain, getting reinterpreted at every layer.

The entire "executive communication" discipline — knowing your audience, not getting too detailed, telling them what they need to know — is training humans to be better encoders for a bad network. We built MBA programs around it.

The Gatekeeping Problem

The biological network doesn't just route information badly. It gates everything:

This isn't meritocracy. It's a social graph with power asymmetries masquerading as a system. Your output matters less than your position in the network and your relationship with the nodes above you.

The Re-Platform

We now have thinking machines. They can:

The org doesn't need layers 1-3. Context onboarding, context syncing, and most machine output can run on thinking machines today. Not theoretically. Right now.

What remains is Layer 4: human thinking and direction. Taste, judgment, vision, novel strategy. The stuff that's genuinely human.

The End State

The org as we know it — the hierarchy, the meetings, the email chains, the politics, the gatekeeping, the forced proximity with people you didn't choose — is not a feature. It's a consequence of running a distributed computing system on biological hardware that was never designed for it.

Re-platform the system onto thinking machines and what's left isn't an org. It's small groups of humans who chose each other, doing the thinking they actually want to do, with agents handling everything underneath.

Work becomes voluntary association, not economic coercion. You work with someone because you like them and their ideas, not because they're assigned to your team. The 80% of life spent on context routing disappears. What remains is the 20% that was always the point.

The org was never the thing. It was the overhead. And we just got machines that can run the overhead.