Industry
What They Will Buy in 2028
To read a market’s future, do not ask what it will do, but what its buyers will buy.
A useful way to reason about a market’s future is not to ask what it will do, but what its buyers will buy. Because the nature of demand always structures, in the end, the nature of supply: a market ends up producing what its buyers want to buy, and not the reverse. Yet the demand of legal organizations is changing more profoundly than one perceives in the current conversations, still centered on products.
The progression is clear for whoever observes it. Legal organizations first bought assistants, when generative AI arrived; they buy today tools specialized by function, as the market fragmented; they will buy, within a few years, no longer assistants nor tools, but semantic layers. This progression is not a theoretical projection, it is already visible in the most advanced purchasing conversations, which have ceased to bear on the performance of an isolated product to bear on something else.
To know where a market is going, do not look at what it sells. Look at what its buyers begin to want to buy.
Demand always precedes supply
It is worth justifying the method before applying it, for it is not obvious. Why reason by demand rather than supply, when it is supply that, apparently, makes the market by launching products? Because supply, in the long run, is only a response to demand. A vendor can launch the product it wants; if it does not match what buyers seek to protect, it does not take hold. Conversely, what buyers durably want always ends up finding, or eliciting, a supply that answers it. Demand is therefore the best predictor of future supply, far more reliable than vendors’ announcements.
This primacy of demand explains why one can read a market’s future in the questions its most advanced buyers ask today. These buyers are not representative of the present market, but they are representative of the market to come: what they demand today, the majority will demand tomorrow, because they simply met earlier the limits the others will meet next. To observe their questions is to observe future demand in advance, before it becomes the majority and structures supply.
This is why the progression of legal purchases, from assistants to tools then to layers, is not a speculation but an extrapolation. It extends a trajectory already drawn by the most mature buyers, whose questions have already left the terrain of the product for that of infrastructure. It is not about guessing what the market will want, but about reading, in those who are ahead, what the others will want when they have traveled the same path.
Questions that have changed in nature
These conversations now bear on questions of another order. What is the underlying architecture? How does the layer hold over time? What becomes of the investment when the model beneath changes? What memory proper to the organization builds in usage? How is governance carried at scale? What trace will allow defending a deliverable three years later? These questions do not resemble those of two years ago: they are infrastructure-purchase questions, not product-purchase questions.
This shift of questions is not anecdotal: it reveals a change in what the buyer seeks to protect. At the start, they sought to obtain a capability, and thus judged on performance: does the tool do well what it promises? Now, they seek to protect an asset over time, and thus judge on architecture, durability, governance: will what I build hold, and will it stay mine? One does not ask the same questions depending on whether one buys a service or builds a patrimony.
This tipping, from service to patrimony, is the real content of the change in demand. To buy a service is to obtain something rendered, consumed, replaced: the question that counts is the quality of the service. To build a patrimony is to accumulate something that remains and appreciates: the question that counts is the solidity of what one accumulates and the guarantee one will stay its owner. Legal organizations pass, in their way of buying AI, from the first logic to the second, and this passage changes everything they expect of a vendor.
One does not ask the same questions depending on whether one buys a service or builds a patrimony.
Three things they will really buy
When this transition of demand is complete, legal organizations will buy three essential things. An operational memory proper to their craft, able to survive changes of vendors and models, and to retain what the organization has decided and learned. It is the first element, and the most structuring, because it is what transforms a scattered usage into an asset that accumulates.
They will buy, second, an orchestration of their workflows that speaks their language and respects their structure. Not a tool that executes an isolated task, but a layer that chains the steps of real work, taking into account the methods, validations, rules proper to the organization. It is what distinguishes an AI that helps on fragments from an AI that carries entire processes, and it is this second form the demand will require.
They will buy, third, an operational governance that makes their usage defensible, traceable, auditable. Not a displayed compliance, but a governance inscribed in the functioning, able to account at any moment for what was produced, by what means, under what control. These three elements, memory, orchestration, governance, taken together, constitute what one calls a semantic layer. And it is exactly this that they will buy, whether they use the word or not.
Memory, orchestration, governance: three distinct purchases today, one single layer tomorrow.
Why one must build before the demand
The stake, for legal organizations, is therefore not to predict what will sell in a few years. It is to understand that they will buy it, and to make now the architectural decisions that will let them arrive at that moment with a layer already constituted, rather than with a stack of tools to reconfigure. The difference between these two situations, when the time comes, will not be caught up in a few months, because a layer is the product of an accumulation time one cannot compress.
It will be objected that anticipating a demand not yet in the majority by several years is a risky bet. It is one, but it is the only one possible for this type of category. A semantic layer is not built in a few months when demand arrives; it demands years of craft depth and context accumulation. Whoever waits for the demand to be there to start arrives structurally too late, not for lack of means, but because the construction time is incompressible.
It is a characteristic proper to infrastructures, and it holds for the buyer as for the builder. The right moment to build an infrastructure is never the one where everyone demands it, because at that moment it is already too late to build it in time. It is the moment, more uncomfortable, where one sees it coming before the others, where demand is still only the fact of the most advanced buyers. To act at that moment is to accept building for a demand not yet there, which is precisely what building an infrastructure has always meant.
The right moment to build an infrastructure is not when everyone demands it. It is before, when you see it coming.
From a capacity buyer to an asset keeper
This change of demand also transforms the very figure of the buyer. Yesterday, the legal AI buyer was a capacity buyer: they evaluated products, compared performances, chose the best tool for a task, in a logic close to buying a service. Tomorrow, they will be an asset keeper: someone whose role is to protect and grow, over time, the intellectual infrastructure of their organization. It is not the same function, and they are not the same decision criteria.
This transformation of the buyer explains the new nature of their questions. A capacity buyer asks “does it work well”; an asset keeper asks “will it hold, will it stay mine, will it appreciate.” The first reasons at the scale of a present need; the second reasons at the scale of the organization’s patrimony over several years. The vendor who can answer only the first’s questions loses the second, because it speaks a language the asset keeper has already left.
One can therefore anticipate that the purchase conversation itself will change interlocutor. As long as one buys a capacity, the decision can stay at the level of the teams that will use it; the moment one protects a long-term asset, the decision rises toward those who answer for the organization’s strategy and patrimony. The change of demand is therefore not only a change of purchased product, it is a change of the level at which the purchase is decided, and thus of the criteria that govern it.
The time lag that defines the category
It is exactly at this future purchase moment that MAX is positioned. The Legal Semantic Layer does not aim to win the debate on the best legal assistant of the moment; it aims to be ready when demand, in a few years, will have migrated toward the category that did not yet really exist three years earlier. It is the right time lag for this type of category: to build before the market names, to be available at the moment it will name.
One must measure how deliberate this lag is. To build a layer today, when the dominant demand still bears on tools, supposes accepting a time where one builds ahead of the market, without the immediate validation that alignment with present demand would provide. It is the entry price of any infrastructure category: it is built in the lag, and those who cannot bear this lag wait, to start, for a moment when the place will already be taken. The lag is not a flaw of the strategy, it is its condition.
The organizations that understand this logic buy, now, the layer of the future, by choosing to build a patrimony rather than consume a service. The others will buy, for two or three more years, the tools of the previous phase, and will find themselves, when the time comes, having to constitute in a hurry a layer the first ones will have had time to build. The demand of 2028 is predictable; what will separate organizations is the moment each will have decided to answer it, early or late.
Tomorrow’s demand is predictable. What will separate organizations is having built before it arrives.
