Two Roads to the Same Place? Haier’s Rendanheyi and Block’s AI-Native Org Experiment

A case note for practitioners asking "what does organizational AI transformation actually look like?"

Why These Two Cases

When people ask for real-world examples of AI-driven organizational redesign, the conversation usually drifts to tooling: which copilot, which platform, which pilot. That misses the more interesting question, which is structural: can technology finally replace the coordination function that middle management has performed for a century?

Two organizations, on opposite sides of the world and starting from opposite ends of the economy, are effectively running the same experiment:

• Haier — a Chinese white-goods manufacturer that has been dismantling its own hierarchy since roughly 2005 through the Rendanheyi ("人单合一", loosely "integrating employee value with user value") model.

• Block — the US fintech founded by Jack Dorsey, which has publicly articulated a vision of removing management layers so that thousands of employees ultimately report to the CEO, with AI systems handling information routing.

They look nothing alike. But they are trying to solve the same problem: how do you flatten an organization without losing the coordination that hierarchy provided?

Case 1: Haier — Twenty Years of Dismantling Hierarchy by Design

The Problem

By the mid-2000s, Haier was already among the world's leading white-goods manufacturers, but it was showing classic large-company pathologies: too many layers, long decision chains, and frontline employees too far removed from the actual user. Founder Zhang Ruimin's diagnosis was blunt: under a bureaucratic pyramid, employees behave as instruction-executing tools rather than autonomous value creators.

The Intervention

Rendanheyi pushed three rights down to the front line: decision rights, hiring rights, and distribution (compensation) rights. The company was broken into small, self-managing micro-enterprises that face the market directly. Structurally, it is often drawn as an inverted triangle: people are pushed toward the customer, and each unit closes its own loop with users.

Evidence It Worked

The most cited external validation is the post-acquisition integration of GE Appliances, where Haier applied the model and — per the source article and Haier's public materials — roughly doubled revenue over five years. That matters because it suggests the model is transferable, not just a culturally specific artifact.

The Problem It Created

Here is the part practitioners should pay closest attention to, because it is the failure mode that every decentralization program eventually hits.

Once you have hundreds of small units all sprinting independently, who reconnects the resources, capabilities, and demand? The source article argues that over-dispersed resource allocation contributed to Haier losing ground to Gree and Midea in core categories such as air conditioning, and cites a persistent cost structure gap (total "four expenses" reportedly above 24%, roughly seven percentage points higher than Midea). I could not independently verify these financial figures — treat them as the source's claim rather than established fact.

Haier's answer was a networked / platform strategy: turn the company into a platform for entrepreneurship rather than a big company — "platform enterprise, entrepreneurial employees, personalized users," supported by mechanisms such as chain-group contracts and internal market pricing.

The honest read: Rendanheyi solved the motivation problem; the networked strategy was an attempt to solve the connection problem. But the connection solution still runs on human contracts, negotiation, and governance rules — which means coordination cost never fully goes away.

Haier's AI Layer Today

Haier is not a "no-AI" case. Its public materials describe:

• HomeGPT — a smart-home vertical domain model.

• Tianzhi Industrial Model — reportedly integrating 4,700+ mechanism models and 200+ expert algorithms.

• Healthcare — automated tumor drug-compounding robots using AI recognition.

• Uhome — driving a global "super agent" for proactive information push and faster decision response.

But the maturity picture is nuanced: most of this is AI applied on top of existing business scenarios. Cross-scenario orchestration, data unification, and seamless human–machine interaction are still evolving.

Case 2: Block — Rebuilding the Org Around Two World Models

The Thesis

Dorsey's argument, articulated with Sequoia's Roelof Botha in From Hierarchy to Intelligence, is historically framed: many companies have tried to escape hierarchy, and nearly all reverted, because nothing existed to replace the up-and-down information transmission that hierarchy provides. AI, he argues, is the first technology that genuinely can.

The Structure

Where Haier is an inverted triangle, Block aspires to a network taken to its limit: in the ideal end state, everyone reports to the CEO, and the AI system handles global information scheduling.

The Enabling Assets

Two internally built "world models" are the foundation:

1. A Company World Model: Because Block has operated remote-first for years, decisions, code, and discussions are natively documented and machine-readable. The system can map the company's operating state in real time. Information that once traveled layer by layer is now carried by the model.

2. A Customer World Model: Built on both consumer- and merchant-side transaction data, constructing a per-customer, per-merchant understanding of financial reality that compounds over time.

What Changes Operationally

Information aggregation, progress tracking, cross-functional alignment, priority sequencing — the coordination work that reportedly consumes ~80% of a middle manager's time — is handed to the model. Frontline employees no longer need a manager to relay decision context; they query the world model directly and decide.

The Open Risk

Removing information routing does not remove the need for judgment. If 6,000 people report to one CEO, every one of them needs strong independent judgment, problem-definition ability, and cross-domain collaboration skill. That capability is unproven at scale, and it is the single biggest unknown in the Block case.

A Useful Frame for Comparing Them

The source article borrows a formula from the "intelligence density" community (attributed to Wang Yuhao, CAIO Alliance):

Where $\text{HI}$ is Human Intelligence, $\text{AI}$ is Artificial Intelligence, and $k$ is the human–machine collaboration coefficient.

The multiplication sign is the whole point. Addition implies division of labor — everyone works separately and outputs are summed. Multiplication implies coupling: if any factor trends toward zero, the whole product trends toward zero.

$k$ is the most overlooked factor. Same people, same model — but whether they can easily find information, connect across systems, and turn experience into reusable skills can change outcomes by an order of magnitude.

So there are only three levers: raise AI capability, raise human judgment, or shorten the distance between the two.

What a Practitioner Should Take From This

The framework's practical value is that the marginal return is highest on your weakest factor.

• For Haier-type organizations (strong culture, deep frontline judgment, legacy systems): building a reusable skill library and investing in AI capability and $k$ will likely yield far more than further strengthening $\text{HI}$, which is already thick.

• For Block-type organizations (strong AI infrastructure, low friction, flat by design): the leverage is in $\text{HI}$ depth — deliberately developing independent judgment, problem framing, and cross-domain collaboration on the front line.

Both are converging on the same design ideal: direct customer contact, flat structure, small autonomous units, zero coordination friction. They differ only in starting point and in which factor they had to earn first.

The likely end state is not "AI replaces humans." It is high-quality human judgment in deep symbiosis with high-performance AI — with the middle coordination layer, not the people, being what gets removed.

Caveats for the Research Record

To ensure transparency regarding evidence quality for research applications:

• Financial and performance figures (the ~24% expense ratio, the seven-point gap vs. Midea, the GEA revenue doubling, the "80% of middle-manager time") come from the source article and Haier's public materials. They should be independently verified before being cited as findings.

• Block's structure is a stated vision and in-progress transformation, not a documented completed end state. "6,000 people reporting to the CEO" is an aspiration described in From Hierarchy to Intelligence, and outcomes are not yet measurable.

• The $\text{HI} \times \text{AI} \times k$ formula is a heuristic framework, not an empirically validated model. It is useful for structuring diagnosis, not for quantitative prediction.

• Source material: Jack Dorsey & Roelof Botha, From Hierarchy to Intelligence; Haier Group public materials on Rendanheyi, networked strategy, and ecosystem brand strategy; Li Xin, 20-Year Retrospective on Rendanheyi; Wang Yuhao, CAIO Alliance "intelligence density."*