Industry
What Firms Really Ask For
Ask the question. No one answers “one more assistant.” What comes up has the shape of an infrastructure.
Ask a legal department or a partner directly: what is your real need in legal AI? You will almost never hear “one more assistant.” It is an answer one does not hear, and its absence is the starting point of everything that follows.
What comes up when you listen to the field
When you dig, what comes up is of an entirely different nature. Finding what was decided on an old matter. Keeping a team’s coherence despite departures. Supervising what juniors produce with AI. Tracing decisions. Holding a methodology without documenting it by hand. Reducing the coordination cost between too many tools. Being able to answer an audit. Not depending on a single vendor. Each of these needs was the subject of one of this pillar’s articles; strung together, they sketch something other than a list of functions.
For none of these needs has the shape of a chatbot. All have the shape of an infrastructure. It is no accident that, asked about their expectations, leaders spontaneously describe properties of holding, continuity, governance, and never generation features. They do not ask for a tool that produces better; they ask for something that holds the whole together, and that something has no name in the vocabulary of assistants.
Tools answer tasks. Organizations seek coherence. It is not the same object.
It is worth pausing on this list, because it has a coherence those who express it do not always perceive. Finding an old decision, holding a team’s coherence, supervising, tracing, governing access, not depending on a vendor: these are not scattered needs, they are the faces of a single need, that of an overall holding the tools, taken one by one, do not provide. Leaders describe, without knowing it, the properties of a layer, because it is a layer they lack, even if they do not have the word.
A task has an end, coherence does not
The nuance is not semantic. A task has a beginning and an end: drafting this clause, summarizing this judgment, finding this precedent. Coherence has no end; it is the property that makes a thousand tasks, handled by twenty people over three years, still tell the same legal story. A tool excellent at the task can let coherence unravel, and that is precisely what happens today in the best-equipped organizations.
This gap explains a paradox the whole market observes without explaining to itself: it has filled with interesting tools and remains, at the same time, deeply unsatisfied. Tools handle tasks; organizations seek an overall holding. And a well-handled task never replaces an absent coherence. One can accumulate the most capable tools and see, nonetheless, the whole come undone, because none of them had the role of holding the whole.
This need will not be solved by addition, and it is the point organizations take longest to admit, because it contradicts their equipment reflex. You add an assistant to an assistant, you do not obtain an infrastructure: you obtain one more assistant, whose marginal benefit decreases and whose coordination cost climbs. Infrastructure is not a particularly successful sum of tools; it is an object of another nature, situated above the tools and holding them together.
You do not add assistants to obtain an infrastructure. You change category.
One can measure this change of category by a simple sign: it changes the decision-maker. An assistant is chosen at the level of the team that uses it; an infrastructure is chosen at the level of the organization that commits to it. As long as one reasons in tools, the decision stays dispersed, tactical, reversible, made by those who execute. The moment one reasons in a layer, the decision rises, becomes strategic, commits the organization for years. It is not only the object bought that changes; it is the level at which one decides, and the duration over which one commits.
This shift, from reasoning in tools to reasoning in a layer, is the true subject of this whole pillar. Each of its articles showed, from a different angle, the same thing: that the problem was never in the quality of generation, always in what surrounds it and is missing. Context that does not fit in a prompt, review made impossible, memory that is lost, method that does not transmit, the absent trace, friction between tools, abandonment midway, ungovernability, dependence on the model: so many manifestations of a single missing layer. Naming them separately was preparing the moment one understands they are but one.
The word “category” is not a dressing. It designates a change in what one buys and in the question one asks. As long as one reasons in tools, one asks “which is the best?”. The moment one reasons in a layer, the question becomes “what holds all this together, and will still hold it in five years?”. It is not a move upmarket, it is another plane, and until one has moved to it, one keeps comparing tools without having asked the question that matters.
Let it be clear: the layer does not make tools useless. The best drafting, review, research tools remain precious. What changes is their status. They cease to be the strategy to become the components of a strategy. One does not choose one’s firm on the brand of its word processor; one will not define one’s AI policy by the list of its assistants. Assistants become features; the layer that holds them becomes the infrastructure, and it is this, now, that deserves the name of strategy.
What MAX builds has a name: the Legal Semantic Layer. A layer above the models, that holds what they do not hold, context, memory, methodology, permissions, governance, the continuity of work. No longer one more product to compare in a list, but an architecture to choose. It is this category that firms describe without naming it when asked for their real need, and it is this category that MAX has chosen to occupy, because that is where, and not in one more assistant, what they truly seek is found.
It remains to draw the practical consequence of all this for a leader who must decide. It holds in an inversion of the question they ask. To stop asking “what is the best tool,” an endless question since the answer changes every quarter, and to ask “what is the layer that will hold all this together, and will still hold it in five years.” This second question does not have the same apparent urgency as the first, but it is the only one whose answer has lasting value. To answer it is to stop chasing tools in order to choose the architecture that holds them.
What firms describe without knowing how to name it is a layer. Not one more assistant.
