Architecture
The Cycle Is Built Above the Models
A software cycle is never defined by its base technology, but by the layer above. The next one will not be either.
Enterprise software cycles are not defined by the base technology that serves as their foundation. They are defined by the layer above that technology, the one that makes the foundation usable at the scale of organizations. This historical regularity deserves recalling, because it is replaying before our eyes with language models, and its reading conditions the right investment decisions. It is constant enough to be relied on, and counterintuitive enough to be forgotten at each cycle.
It is simply put: one names a cycle after its base technology, but one lives it after the layer above. The base technology gives its name to the era, because it is the visible novelty, the one talked about. But the value, the real usage, the revenue concentrate one notch above, in the layer that transforms this raw novelty into something organizations can really employ. The name and the value are not on the same floor, and it is the source of most investment errors.
One names a cycle after its base technology. One lives it after the layer above.
Four cycles, one same rule
One can verify this rule on the successive cycles of enterprise software, and it never weakens. The cycle associated with the great central computers was not, in reality, that of those machines, but that of the integrated management software built above them, which captured the value. The next cycle was not that of relational databases, a decisive base technology nonetheless, but that of the analytics and semantic layer built on top to make them usable by the crafts.
What follows confirms the rule without exception. The cycle associated with programming interfaces was not theirs, but that of the online platforms that exploited them to deliver entire applications. The cycle associated with the cloud was not that of remote hosting in itself, but that of the governance, security and productivity platforms built above the cloud. Each time, the base technology gave its name to the cycle, and the layer above captured the value. The regularity is too constant to be chance.
One can wonder about the reason for this regularity, for understanding why it holds allows relying on it for the future. The base technology of a cycle is, by nature, generic and raw: it opens possibilities, but it does not organize them. It is organizations that need these possibilities ordered, governed, linked to their processes, and it is precisely this work of ordering that the layer above performs. The base technology makes the headline; the layer above makes the revenue. This is why one names cycles after the first and lives them after the second.
The foundation makes the headline. The layer above makes the revenue.
Why the next cycle will not be the models’
The cycle opening now will not be an exception to this rule. It will not be the cycle of language models, despite all the noise they make today, but that of the layer built above them to make them usable in organizations. Models are the base technology of this cycle: decisive, necessary, but generic and raw, exactly as databases or the cloud were in their time. And as at each cycle, it is not the foundation that will define the era, it is what will be built on top.
This layer has no widely consensual name yet, but its contours are already visible: it is semantic, orchestrated, governed, persistent, integrated with existing tools, specialized by major domain of expertise. It is this that will define the decade of enterprise software opening. Not the models, not the copilots, not the assistants, which are the visible surface of the cycle, not its core. The strategic consequence is simple to state and heavy to carry: the players building surface tools today operate on what characterizes the end of the previous cycle, not the start of the new one.
One must measure how unsettling this reading is for the present. All the noise, all the money, all the attention go today to the base technology, the models, exactly as, at each past cycle, they went to the foundation of the moment before shifting to the layer above. Those who, at each cycle, invested in the foundation alone bet on what was going to commoditize; those who invested in the layer above bet on what was going to capture the value. Nothing indicates this cycle will reward the inverse error of the previous ones.
The next cycle will not be the models’. It will be that of what is built on top of them.
The trap of naming the cycle after its foundation
One must understand why this error repeats at each cycle, despite its retrospective obviousness. The trap comes from the foundation being what strikes the imagination. A new base technology is spectacular: it makes possible what was not, it changes the rules, it occupies conversations. The layer above, beside it, seems modest: it only organizes, links, governs what the foundation made possible. It is human to believe value is on the side of what impresses, and it is precisely this belief that misleads investment, cycle after cycle, toward the foundation rather than the layer.
This bias in favor of the spectacular has a predictable consequence: at the start of each cycle, the bulk of capital and talent goes to the foundation, and it takes several years for the market to understand that value has settled elsewhere. This time lag is an opportunity for whoever understood it early. While the crowd invests in the impressive foundation, whoever read the historical regularity builds the discreet layer, and takes, without competition, a position the others will covet only too late, when it is already occupied.
One can therefore read the present of AI as a classic moment of this bias. Models impress, occupy all the space, attract the bulk of attention and capital; the layer above, less spectacular, builds itself in relative discretion. If the historical regularity holds, and nothing indicates it will yield, this moment is exactly the one where one must resist the appeal of the foundation to build the layer. Those who yield to the appeal of the spectacular redo the error of all previous cycles; those who resist it draw the lesson.
The layer above is also the most defensible
One must add that this position is not only more precious, it is also more defensible, and the two qualities are linked. The base technology of a cycle is, almost by definition, available to all: anyone can use the central computer, the database, the cloud, the model. This general availability is what makes its diffusion power, but it is also what forbids making it a durable advantage: what everyone can use distinguishes no one.
The layer above, on the contrary, accumulates each organization’s own context, and this context does not transfer. Whoever builds on the foundation alone builds on common ground, where competition is immediate and the advantage precarious; whoever builds the layer above builds on ground that becomes, over time, their own. The choice of where one builds is therefore also a choice of defensibility: on the foundation, one is one among others; on the layer, one accumulates something the others cannot take back.
This defensibility by accumulation explains why the layer above, at each cycle, ended up worth more than the foundation. Not because it was more impressive, it rarely was, but because it capitalized an own context the foundation, generic by nature, could not capitalize. Value always goes, in the end, where something accumulates, and nothing accumulates in a foundation shared by all. It is this logic of accumulation, more than performance, that decides where a cycle’s durable value resides.
The foundation is open to all. The layer above is what becomes, over time, truly yours.
Choosing the floor on which to build
It is precisely MAX’s bet in building the Legal Semantic Layer: not a surface tool meant to profit from the enthusiasm phase for models, but an infrastructure layer conceived to become, in its domain, one of the structuring components of the next cycle of legal enterprise software. The strategic position to take is not in the foundation, it is in the layer above, and it is built today, in the architectural choices of the coming quarters.
One must, finally, draw the practical consequence of this regularity for whoever must decide now. Since the cycle will be defined by the layer above, the useful question is not “which model to adopt,” which bears on the foundation, but “which layer to build,” which bears on what will define the cycle. An organization that structures its thinking around the model reasons at the level that will give the cycle its name, but not its value; an organization that structures it around the layer reasons at the level where value will actually reside. It is not a nuance, it is the choice of the floor on which one decides to build.
This choice of floor is made early, and that is what makes it decisive. Building the layer above takes time, because it accumulates in usage; one cannot build it at the last moment, when the foundation has commoditized and everyone has understood where the value was. At that moment, those who started early will have a lead made of accumulated context, which the latecomers will not be able to buy back. The strategic position is never in the foundation, it is in the layer above, and it is taken at the moment the foundation still occupies all the attention.
The strategic position is never in the foundation. It is in the layer above, and it is taken early.
