Layers vs. silos: why reorgs don't modernize

The organizational reflex

Every time an organization feels AI is leaving it behind, the same thing happens: someone convenes a restructuring. An "AI office" is created, a lead is named, a new org chart is drawn with a new box connected by dotted lines to the old boxes. Six months later, the AI office reports progress, the old boxes keep operating exactly as before, and the organization still can't answer the question that started all this: which part of what we do today can a model do, and which part can no model touch without the organization losing something it didn't know it had?

The reflex has an understandable logic. An org chart is the only change tool most executive teams handle fluently. When something doesn't work, you reorganize; when something new matters, you create a box for it. But an org chart describes who reports to whom, not which layer each function operates in — and that is precisely the distinction AI modernization demands and that no redesign of boxes can produce.

Layers, not silos

A silo is a division by domain: finance, operations, product, legal. Each silo has its own decision chain and its own execution chain, both mixed inside the same box. When a silo is reorganized, people and reporting lines move, but the mix of decision and execution inside each box doesn't change — it simply shifts into a box with another name.

A layer is different: it cuts across every silo at once. An organization's decision layer — what is pursued, what is considered valuable, what risks are accepted, who holds final authority — doesn't live in a department. It lives scattered across the executive committee, the written policies, and the unwritten precedents everyone knows but no one documented. The execution layer — how the work gets done, with which tools, on which infrastructure, with which vendor — also cuts across every silo: finance executes on a contracted ERP, product executes on a contracted cloud stack, legal executes on contracted document-management software. Different boxes, same layer pattern.

AI modernization doesn't ask a silo "how do you reorganize?" It asks a layer: "is this function execution that can be modernized with a new tool, or is it decision that the organization can't afford to delegate without losing control of what actually matters to it?" That question has no departmental version. It has a layer version, and it cuts across the entire org chart without respecting a single dotted line.

What actually moves

This doesn't mean the org chart is irrelevant — it means reorganizing it is a low-leverage intervention when the real problem is one of layers. Three concrete moves pay off more than a restructuring:

First, audit by layer before auditing by department. Instead of asking "what does the operations team do?", ask "of everything currently executed across the whole organization, what portion is replaceable commodity and what portion is the real competitive advantage, regardless of which box each lives in." That audit cuts across silos and tends to reveal that the real competitive advantage is spread in small fragments across departments that, seen separately, looked entirely replaceable.

Second, place decision authority where the knowledge of the risk is, not where the highest available title is. When an execution layer is modernized with a new tool, someone has to decide what reversibility to demand of the contract — and that decision should be made by whoever understands the real technical dependency, not necessarily whoever signs the budget.

Third, measure modernization at the layer, not at the department. If the goal is "cut a process's time by 40%," the right question isn't which department achieved it, but whether that reduction came from delegating execution reversibly or from quietly ceding a portion of the decision layer in exchange for speed. The first is real modernization. The second is a surrender disguised as efficiency, and the new org chart won't show it in any quarterly report.

Reorganizing boxes feels like progress because it's visible, fast, and easy to announce. Redesigning layers is slower, less photogenic, and it's the only thing that actually changes what the organization controls when the next model — cheaper, more capable, from another vendor — shows up in eighteen months.