Practice

The Hidden Economics of Friction

The cost of seven AI tools is not in their seven subscriptions. It is in everything they force you to do to make them hold together.

In eighteen months, most serious legal organizations have equipped themselves. One tool for retrieval, another for drafting, a third for review, a copilot for research, an assistant for summaries, a platform for KM. Each chosen for good reasons, each good at its task. And yet, the more you add, the less the whole behaves like a coherent system.

One instinctively measures the cost of this equipment by the sum of the subscriptions. That is the visible part, the one in the budget, the one negotiated and compared. It is also the least important. The real cost is elsewhere: it is diffuse, it appears on no invoice, and it is probably several times greater than the sum of the licenses. To see it, one must stop counting what is paid to vendors and start counting what is paid internally to make what was bought from them work.

The subscription is the cost you see. It is almost never the cost that matters.

The cost with no budget line

Let us look closely at what a stack of tools that ignore one another produces. The drafting tool ignores what the review tool has already commented on. The copilot ignores that KM already contains the firm’s position on the question. The assistant ignores that retrieval brought back the right precedent the day before yesterday. The same question is asked three times of three tools that return three slightly different answers, and it is the user who must arbitrate between them. Legal work insensibly turns into brokerage between artificial intelligences that ignore each other.

Each of these frictions costs a few minutes. Reformulating a request in another interface. Checking that two tools do not contradict each other. Rebuilding by hand the context each lost on its side. A few minutes, but dozens of times a day, across dozens of people. The initial promise of time saved dissolves in this permanent coordination effort, and the net benefit becomes so hard to isolate that many deployments struggle, months later, to prove any return at all.

One must name precisely the nature of this cost, for that is what makes it so easy to ignore. It is not a purchase cost, one-off and negotiated; it is a usage cost, recurring and invisible, paid not in euros handed to a vendor, but in minutes taken from professionals’ time. Yet a legal professional’s time is the organization’s most expensive resource. To pay this cost in expert time is to pay it at the highest rate there is, while having the illusion of paying nothing, since no invoice materializes it.

The subscription is the cost you see. Friction is the cost that eats the benefit.

What seven tools do to the work itself

Beyond the lost time, one must see what this dispersion does to the nature of the work. A professional juggling seven tools does not merely lose minutes; they change posture. They cease to be the one who reasons to become the one who coordinates, the one who checks coherence between competing outputs, the one who carries context from one window to the next. This shift is insidious, because it does not feel like a degradation: each coordination gesture seems necessary, almost professional. But strung together, these gestures move the expert’s work from substance to logistics.

This transformation has a cost measured not in hours but in quality of attention. Legal reasoning demands continuity, sustained concentration on a single problem. Coordination between tools fragments it endlessly: you leave an analysis to go look elsewhere, you come back, you have lost the thread, you pick it up again. This permanent fragmentation wears out a resource rarer still than time, deep attention, and it wears it out precisely in those from whom deep work was expected.

One then understands that the cost of the seven tools is not only quantitative. It is also qualitative: it lowers the level at which the expert can work, by constantly returning them to stitching tasks that require none of their skills. A firm that equips its best people with seven uncoordinated tools obtains, without wanting to, that its best people spend part of their time doing work anyone could do, at the expense of the work no one else can do.

The calculation no budget makes

Let us do the math, even roughly, for the order of magnitude is eloquent. Suppose each professional loses twenty minutes a day to coordination frictions between tools, a prudent assumption given what one observes. That represents a little over seventy hours a year per person. On a team of thirty, one exceeds two thousand annual hours: the equivalent of a full-time position devoted solely to making tools that do not talk to each other converse.

This expense has a remarkable property: it is never decided. No one, in the organization, ever approved the creation of a full-time position for manual coordination between tools. It appears in no budget, is the subject of no review, appears on no dashboard. It is simply absorbed, in silence, spread in thin slices across the schedules of the organization’s most expensive people, until it becomes invisible by dint of being diffuse.

And this calculation is still optimistic, because it counts only the time directly lost. It does not count the cost of the errors born of missed coordination: the contradictory position taken because one tool ignored what another knew, the verification omitted because each thought the other had done it, the rework because context was lost between two steps. These costs, rarer but far heavier, add to the lost time, and they are even less traceable, because they are attributed to individual faults rather than to the absence of a layer that made them probable.

You budget licenses. You absorb, without seeing it, the equivalent of a full-time coordination job.

Why consolidating from below does not work

The natural reaction, faced with this observation, is to want to reduce the number of tools, or to replace several of them with a single, more complete one. But this consolidation from below quickly hits a structural limit. No isolated tool is best everywhere, and the model market evolves too fast for a single vendor to stay durably ahead on all tasks. To want to do everything with one tool is to accept being average everywhere in exchange for a little coherence. It is a bad bargain.

There is, moreover, a deep reason for this impasse. Reducing the number of tools does not remove the need for coherence; it displaces it inside the single tool, which will then have to be good at very different tasks, which none succeeds at durably. One trades a coherence problem between tools for a mediocrity problem inside a tool. The need for articulation does not disappear because one reduces the number of pieces; it recomposes otherwise, and often at a higher cost, because one gave up the best of each tool without solving what was problematic.

Coherence is therefore not won by removing tools, but by capping them. By placing above them a layer that knows what each does, maintains a shared context, chains the steps, and brings coherence not as a permanent effort of the user, but as a property of the system. The difference is the one that separates a team where everyone must constantly coordinate with everyone else, from a team where someone holds the thread of the whole: in the first case, coordination is a burden borne by all; in the second, it is ensured at one point, and frees all the others.

The accumulation of tools is not a strategy. It is the symptom of a missing layer.

The invisible cost always ends up showing

This is the role MAX occupies: not an eighth tool in the stack, but the layer that makes the seven others coherent with each other. Not to replace them, but to remove the hidden cost they impose when they work while ignoring one another. The value of such a layer is measured not by what it produces anew, but by what it makes disappear: the daily twenty minutes, the contradictory answers to arbitrate, the context to rebuild, the full-time position no one had decided to create.

This invisible cost has a reassuring feature: by dint of being paid, it ends up showing. An organization can long ignore that it absorbs the equivalent of a full-time job in coordination, but it ends up finding that its promised productivity gains do not materialize, that its teams are tired, that the early enthusiasm has given way to a diffuse weariness. These are the symptoms of the hidden cost surfacing. And the day an organization puts a name on what it was paying without seeing, the question is no longer to add a tool, but to lay the layer that was missing.

It is on this layer, far more than on the next fashionable tool, that the legal AI market will reorganize in the coming years. The organizations that have understood it will stop counting their tools and start counting what those tools cost them in coordination, and they will seek not the eighth tool, but the layer that makes the first seven coherent at last.

The next gain will not come from one more tool. It will come from the layer that removes the cost of the tools you already have.