From application to operating system

Watch where the capital is going. AI-native ERP cores have raised nine-figure war chests in under two years — and every one of them is finance-first: ledger, close, revenue. Meanwhile the most interesting move in the category was made by a company that chose not to be a core at all: an AI-native inventory layer that plugs into those finance cores. The unbundling of the ERP has begun, and the market is already sorting itself into cores and modules.

What nobody has claimed is the seat we care about: the cross-functional operational core for companies that make and move physical goods — and the marketplace above it. Items, BOMs, inventory, capacity, orders: the operational truth that finance-first systems treat as someone else's problem.

Here is the trap in every "platform" pitch, including ours if we are careless: no great platform started as a platform. Windows inherited an application monopoly before anyone wrote Windows apps. The iPhone sold for a year before the App Store existed. Salesforce was a CRM for five years before AppExchange. Shopify was checkout software for a snowboard shop. Platforms are not founded — they are earned, by an application so useful it accumulates the two assets a platform needs: customers, and data custody.

So we start with an application. Governed decision loops for Consumer Goods supply chains — replenishment exceptions first, because they fire daily, their outcomes are measurable, their blast radius is bounded, and the data already sits in the ERP. The visible job of the wedge is to close decisions under governance. Its hidden job is to earn custody of operational master data, one connected system at a time.

The wedge earns the kernel. Running the loops forces the unglamorous work — connecting, cleaning, and contracting the operational truth. Done once, properly, that becomes a product in its own right: a semantic core with identity, permission scopes, versioned data contracts, and an event bus, exposed through open protocols. Your company's API. Every AI tool you adopt connects through one governed interface instead of forty brittle point integrations. We adopt open standards for the wire; the differentiation is the semantic model and the accumulated outcome data, not a proprietary format.

The kernel earns the store. Once modules read and write through declared data scopes — app-style permissions, enforced by the core — switching providers becomes configuration, not migration. Five MRP vendors can compete on capability, price, and service. And because the platform routes the flows and holds ground truth, it can do what no app store has ever done: score modules on measured outcomes — forecast fit, plan adherence, SLA, cost per decision — with a published methodology and an appeal process. Reviews rate software. An operating layer can measure it.

The store earns the brain: orchestration across modules toward one enterprise objective, under the same decision rights and audit envelope we ship in the first loop. Nothing from the wedge is wasted — the decision layer becomes the scheduler; governance becomes the trust infrastructure of the marketplace.

We hold ourselves to three rules on the way. Honest sequencing: the first years are application and core revenue; the take rate is the prize, not the plan. Neutrality: our own modules use only the public APIs every third party gets. And honest scope, always: we are early-stage, and we will keep saying so until the measured results say otherwise.