The New Triumvirate: The Three Roles That Survive

In February 2026, Jack Dorsey cut roughly 4,000 people from Block — about 40% of the workforce, taking headcount from 10,000 to 6,000. The stock rose 22% on the announcement. This wasn't a cost-cutting exercise dressed in ideology; Block projected $12.2 billion in gross profit for the year, 18% growth. Dorsey's explanation was almost philosophical: corporate hierarchy was always a solution to a coordination problem. AI now solves that problem. So the hierarchy becomes redundant.

Two months later, Dorsey and Sequoia's Roelof Botha published the theory behind the cut, "From Hierarchy to Intelligence." And here's the part worth pausing on. The essay named three surviving human roles. They map, almost exactly, onto a structure derived independently from first principles. Two analyses of what human work looks like after AI absorbs coordination — one from a publicly traded fintech, one from architectural reasoning — converged on the same three categories.

That convergence is the strongest evidence I know that the New Triumvirate isn't a tidy abstraction. It's the shape organizations settle into once execution becomes machine territory.

The mistake: assuming the pyramid just gets shorter

The conventional read on AI and org design is that the pyramid compresses. You keep the same shape, lay off some middle, and run leaner. It's an intuitive picture, and it's wrong in an important way.

Compressing a pyramid assumes the remaining roles are smaller versions of the old ones — fewer managers managing fewer people. But a manager's core function was routing information through an organization too large for any one person to oversee. That function doesn't shrink when agents take it over. It moves. The work that survives isn't "management, but less of it." It's a different set of jobs entirely, defined by what coordination machinery can never supply: chosen purpose, contextual judgment, transmitted craft.

Steel-manning the compression view: it's not crazy. Plenty of firms are getting real, if modest, gains by bolting agents onto existing structures. The trouble is those gains plateau. An NBER survey of nearly 6,000 executives found over 80% of companies implementing AI reported no discernible impact on employment or productivity. That's not a technology failure. It's an architecture failure — leaving the pyramid intact and wondering why the magic didn't happen.

The reframe: three roles, not a shrunken hierarchy

What AI displaces is execution, coordination, routine cognition. What remains divides cleanly into three.

The Architect sets strategic intent — choosing which objectives merit optimizing — and designs the cognitive architecture that executes it. Technology magnifies intent but never originates purpose. AI optimizes a defined objective function; it cannot ask why this, and not something better? The Architect asks first, then translates the answer into executable systems.

The Guardian keeps judgment. When agent swarms escalate ambiguous, high-stakes, or ethically charged decisions, the Guardian makes the call — not mainly for output quality, but for alignment with purpose: is this consistent with what we said we were building?

The Force Multiplier transmits taste by doing the work alongside others, building team capacity through craft rather than management. The role has an older name in practice: Player-Coach.

Figure: Three roles span the complete domain of human agency after AI absorbs coordination — add a fourth and you get redundancy; drop to two and something vital goes unassigned.

Why exactly three

The number isn't arbitrary, and resisting both four and two is the test of whether the framework is real.

Strategic direction requires moral imagination to choose worthy ends and architectural skill to execute them — that's the Architect. Contextual judgment needs institutional accountability to navigate ambiguity algorithms can't resolve — that's the Guardian. Craft transmission demands cultural embeddedness and pedagogical presence no rubric captures — that's the Force Multiplier. These three span the domain. Add a fourth role and you introduce overlap. Cut to two and something essential goes unowned.

Watch how they operate as a system rather than a list. When a Force Multiplier at a fashion platform senses a recommendation engine drifting into monotony, she works with the Architect to redesign the detection architecture and reframes what the platform optimizes for. When a Guardian at an energy company faces a recommendation to abandon its founding mission for profit, resolving it takes all three perspectives — the numbers (Force Multiplier), the architecture cost (Architect), the purpose (Guardian). Each role stays distinct. Every hard call needs all of them.

The mechanism that makes three enough

The roles don't shrink because people became superhuman. They shrink because the organizational plumbing that used to require people now runs on agents.

Block made this literal. It deployed two AI "world models" — an internal one aggregating code, decisions, workflows, and performance metrics into a live picture of operations, and a customer model built from Cash App and Square transaction data. Together they replace what middle management used to provide: context, coordination, institutional memory. Block's Individual Contributors, Directly Responsible Individuals, and Player-Coaches operate with full organizational intelligence available directly, not filtered through twelve reporting layers.

The AI-Born cortex works the same way. The Architect doesn't need a project manager to understand system state — the VP-Agent provides it. The Guardian doesn't need an operations committee to know what agents are doing — the audit trail is continuous and searchable. The Force Multiplier doesn't need a structured onboarding program — she transmits craft by working.

There's an honest corrective in Dorsey's own data. Block reports roughly 90% of its code is now authored by its AI agent, yet employees told journalists about 95% of AI-generated code changes still require human modification, and that AI can't lead in regulated domains like banking. That's not a contradiction of the thesis. It's the thesis. The human modifies, validates, guides, teaches; the agent executes at volume. That division is the three roles.

What to do with this

1. Map your survivors by role, not by rank. For each function, ask which of the three categories it belongs to. Work that fits none of them is probably coordination overhead that agents will absorb.
2. Give the Guardian structural independence. No P&L, no reporting line into the chain it reviews, explicit authority to escalate without consequence. Three powerful people without independent checks is a governance crisis waiting for a trigger.
3. Build the world models before you cut the layers. Block could remove middle management because the AI substitutes for what those layers did. Remove the layer without the substitute and you get chaos, not leverage.
4. Encode joint purpose as seriously as the technical architecture. Roman triumvirates collapsed into civil war when the shared mission dissolved. A three-person cortex without a transcending purpose is fragile in the same way.

The principle

The org chart of the AI-Born firm isn't the old pyramid with floors removed. It's three load-bearing roles — intent, judgment, craft — sitting on a machine core that does the rest. You can derive that from first principles, or you can arrive at it the way Dorsey did, by running a company through the transition and watching what's left standing. Both roads end in the same place. When two independent paths converge on three, it's worth treating three as the destination.