Industry

You Are Asking the 2024 Question

“Which AI to buy” was a good question. It has stopped being one. And to keep asking it is already to buy the wrong AI for 2027.

For two years, a single question occupied the AI committees of legal organizations: which vendor, which model, which platform. Demos followed one another. Pilots multiplied. The question came back, identical, every quarter. This question was useful; it allowed exploring new territory, understanding what the tools could do, sorting the serious from the chatter. It no longer is, and it is the kind of tipping one does not notice at the time, because a question that was long the right one keeps seeming relevant out of habit, well after ceasing to be.

What changed is not the quality of the tools, it is the nature of the problem. Serious organizations today have several tools that each do what they promised. The lack is no longer in selection; it is in articulation. One has solved “which tool” and discovered, in doing so, a problem of another order: how to make these tools work together, without multiplying complexity, without breaking governance, without depending on a vendor. It is a question the first did not let one see, and one meets only after answering the first.

Choosing your tool well is useless if the problem is no longer the choice of tool.

A catalog problem, or an architecture problem

The difference between the old question and the new is not rhetorical, it is of nature. “Which tool to buy” is a catalog problem: it is solved by choosing better from a list, comparing sheets, weighing features. “How do these tools hold together” is an architecture problem: it is solved by changing the plan, no longer which element to take, but how the elements articulate. These are two distinct registers of thought, and one does not reduce to the other.

This distinction has a formidable practical consequence. As long as one treats an architecture problem as a catalog problem, one buys better and solves less. One accumulates good individual choices that produce a bad collective result, because the quality of each piece says nothing of the coherence of the whole. An organization can have selected, one by one, the best tools on the market, and find itself with a system that does not hold, precisely because it answered the wrong question with excellence.

One can measure the gap by the way the answer is built. A catalog problem is handled internally, by comparison, and settled in a purchasing meeting; it commits a budget and closes. An architecture problem does not close at purchase: it commits a way of holding things together over time, a governance, a trajectory. To confuse the two is to believe one has finished when one has signed, when one has not even begun to address the real subject, which was not in the catalog.

A catalog problem is solved by choosing better. An architecture problem is solved by changing the plan.

There is a reason this confusion persists, and it lies in the catalog being reassuring. Choosing from a list is a familiar, bounded exercise that produces a clear decision and a sense of mastery: one compared, one decided, one has an owner and an invoice. The architecture problem, for its part, is uncomfortable: it has no clear end, it belongs to no vendor, it forces reasoning about relations rather than objects. Faced with this discomfort, the temptation is strong to return to the catalog, where at least the questions have simple answers, even if they are not the right questions.

The market has no organized answer yet

This is where the market gets stuck, and it must be said clearly. Vendors offer their tool; none, or almost none, offers the layer that would make the tools coherent. The architecture question is posed by organizations; the offering, for its part, is still structured as a catalog. There is a gap between the maturity of demand, which has understood it has an articulation problem, and the maturity of supply, which keeps selling pieces where it is asked for an overall holding.

One might think the answer is to choose a single platform that does everything. But that is trading one risk for another: fragmentation for dependence. A platform that does everything is a single vendor, a single pace of innovation, a single point of failure, and the impossibility of benefiting from the best model of the moment when it appears elsewhere. Lock-in with a single vendor is not the exit from the architecture problem; it is a degraded form of it, where one has solved coherence by giving up freedom.

The real answer is of another nature: a layer above the existing tools, internal or external, commercial or open-source, that ensures what none holds alone, the memory of matters, the methodology, the orchestration, the permissions, the traceability, while keeping the tools interchangeable. The right layer does not lock in, it frees: it makes the organization owner of its coherence without making it captive to a vendor. It is this category that MAX builds, not one more tool to compare, but a Legal Semantic Layer that transforms a stack of tools into a unified legal environment.

The delay in changing the question is paid dearly, because it is cumulative. Each month spent stacking tools with no layer above is integration debt accumulating: more junctions to hold, more scattered context, more governance to catch up on the day one gets to it. The organization that asks the right question early builds on a sound base; the one that asks it late will first have to undo before being able to build. The debate has already moved to the layer above; the only question left is who understood it in time, and who keeps optimizing a question the market has left behind.

One must be precise about what “changing the question” asks of an organization, for it is not a mere adjustment of vocabulary. It is shifting the center of gravity of the decision. As long as one asks “which to buy,” the decision belongs to those who compare products, and is made at the rhythm of demos. The moment one asks “how does all this hold together,” the decision rises toward those who answer for coherence and governance, and is made at the scale of the organization, for years. It is not the same decision, not the same decision-makers, and not the same horizon. To change the question is to change the level of decision.

To keep asking the 2024 question is to guarantee having the wrong AI in 2027.