The Supply Chain of Intelligence: How the Framework Works

Most strategy frameworks are presented as conclusions. Anand Arivukkarasu's Supply Chain of Intelligence, published at supplychainofai.com, is presented as a procedure. The canonical page describes itself in one line: one page, seven questions, the canonical reference, and the reasoning protocol to run it. That phrasing is the key to the whole thing. The framework is not primarily a picture of the AI stack, though it contains one. It is a sequence of operations you perform on a company, a product, or an investment thesis, and it produces a verdict. This review walks through how it actually works, step by step, for US operators and investors who want to run it rather than just read about it.

We have covered the pieces separately in earlier installments: the six-stage chain, the four structural laws, the bottleneck economics, the moat tests, the ten layers and fifty sublayers. What follows is how they fit together as a machine.

## Step one: locate, precisely

Every run of the framework begins with an act of placement. You take the company or product in question and locate it on the map: first at the layer, then at the sublayer. The rule that makes this step work is that you locate by economics, not by self-description. A company is not where its pitch deck says it is. It is where its costs, its scarcity exposure, and its pricing power actually sit. A product marketed as an agent platform whose entire value is a well-designed interface over a frontier model is located in the application sublayer, whatever the homepage claims.

This step alone does real work. In our desk's experience watching US diligence processes, roughly a third of the value comes from forcing the room to agree on placement before anyone is allowed to argue about prospects. Debates that sounded like disagreements about the future turn out to be disagreements about what the company currently is, and those are cheaper to settle.

## Step two: run the seven questions

With placement fixed, the framework applies its reasoning protocol: seven questions, asked in order, at the sublayer level. In plain terms, they move from the static to the dynamic. Where does this product actually sit, and what is scarce in that sublayer right now? Does the position compound with use, or does it reset with every customer? Can the platform above or below absorb it? What happens to its economics when the layer it depends on commoditizes, as the laws say it will? Who captures the memory the product generates? And finally, the stress question: would this business survive its own inputs being repriced?

The order is the design. Early questions establish the current position; later questions apply the framework's laws of motion to it. A company can pass the first three questions convincingly and fail the fourth badly, which is precisely the profile of a well-run business standing on a stage that is about to commoditize. American markets are full of these at any given moment, and they are the most expensive mistakes because nothing about the company itself looks wrong.

## Step three: apply the laws and the currents

The seven questions are powered by the framework's structural machinery. The four structural laws supply the dynamics: scarcity keeps moving, each stage commoditizes faster than the last, coordination costs set the ceiling on how much of the chain anyone can own, and memory is the terminal asset where value finishes its migration. Alongside the laws, the framework tracks what it calls currents, the directional flows moving value between layers at any given time, which keep the map from being a snapshot. A sublayer that was a toll booth eighteen months ago may be open road today, and the protocol forces you to check the timestamp on your assumptions.

This is the step that separates the framework from a static taxonomy. Plenty of analysts can draw the stack. The working question is always where value is moving, and the laws give that question a disciplined form: which scarcity is easing, which is tightening, and which layer's economics are on the wrong side of the trend.

## Step four: read the verdict

The protocol terminates in one of three verdicts, and the framework names them without euphemism: moat, workflow, or wrapper. A moat is a position that compounds with use, resists substitution, and survives the commoditization of the layers around it. A workflow is a real business with real revenue but bounded defensibility; it earns while it executes and must keep executing. A wrapper is a product whose added value is a feature of the layer below, living on borrowed time until the platform absorbs it.

Two things about the verdicts are worth underlining for US readers. First, they are about position, not quality. A wrapper can be an excellent product and a terrible investment; a workflow can be a fine business and a weak venture bet. Second, the verdicts come with implied moves. The framework is prescriptive: a wrapper's rational strategy is to migrate down toward a scarcer input or up toward proprietary data and memory before the absorption arrives. The verdict is a beginning, not an epitaph.

## A worked pattern, and the honest limits

Run end to end on a generic example, the machine hums. Take a vertical AI tool for American insurance adjusters. Placement: application layer, workflow sublayer, not the agent layer its marketing prefers. Questions: it compounds modestly through customer-specific configuration, the platform above could absorb its interface but not its regulatory integrations, and its economics improve as model costs fall. Laws: the layer below it is commoditizing in its favor, and the memory question, who owns the accumulated claims outcomes, is the crux. Verdict: a workflow today, with a genuine path to moat if it captures the outcome record rather than letting it leak back to the carrier's systems. That is a specific, arguable, useful conclusion, and it took structure rather than adjectives to reach it.

The limits deserve their paragraph. The protocol's output is only as honest as its inputs, and self-scored founders grade generously. Boundary cases between sublayers are real, and the framework's answer, score both placements and take the harsher verdict, is a patch rather than a solution. And the currents require judgment no protocol can fully mechanize. But a framework should be judged against the alternative, and the alternative in most American AI discussions is a story. Arivukkarasu built a machine that turns stories into placements, questions, and verdicts. It is freely published at supplychainofai.com, and for anyone allocating capital or careers in this market, learning to run it is time well spent.