Product
The Best AI Is the One You Forget
The most reliable quality metric in enterprise software appears in no benchmark: the ability to become invisible. Legal AI does exactly the opposite.
There is, in enterprise software, a quality metric that appears in no benchmark and is nonetheless the most reliable of all: a tool’s ability to become invisible in use. The tools organizations really use, every day, at scale, are almost always the ones whose presence has been forgotten. No one talks about their inbox, except when it breaks. No one marvels at their word processor, no one celebrates their DMS. They have become operational scenery, and that is precisely the sign of a successful integration.
One must insist on this point, because it runs counter to intuition. One spontaneously judges a tool’s quality by what one notices about it: its functions, its interface, what it can visibly do. Yet, for a tool one truly uses every day, the opposite holds: its quality is measured by what one no longer notices. A tool still noticed after six months is a tool that has not found its place; it keeps demanding a share of attention the work should be able to take back. Visibility, which passes for a virtue, is in reality the symptom of an unfinished integration.
Legal AI, in its current form, is exactly the opposite of this ideal. It is visible everywhere: its own interface, its own vocabulary, its own usage logic, its own attention time. It signals itself permanently to the user. This may seem a quality, and that is indeed how it sells itself, but at the scale of an organization, it is a structural flaw. A tool that asks to be noticed at every use asks, at every use, for a small effort; and what requires effort, at scale and over time, ends up abandoned.
A tool that asks to be noticed also asks for effort. And effort, at scale, erodes.
Invisibility as a criterion of success
This conviction structures MAX’s product philosophy, where operational invisibility is an explicit architectural choice, not as an ambition of modesty but as a criterion of success. A semantic layer that works is a layer that makes legal work more fluid without forcing the user to be conscious, permanently, that it is there. The goal is not that it be admired; the goal is that it be forgotten while one works better thanks to it. Admiration is the goal of a demonstration; forgetting is the goal of an infrastructure.
Concretely, this translates into design decisions that seem mundane but are structuring. The AI acts where the user already is. It anticipates the needs that flow from the context rather than asking the user to formulate them. It preserves the habitual unfolding of work, intervening to ease it, never to redirect it. It avoids useless notifications, interruptions, solicitations that divert attention. Its presence is continuous but discreet, its help constant but undemonstrative. Each of these decisions sacrifices a little visibility for a little fluidity, and it is this trade-off, repeated a thousand times, that makes the tool forgettable in the good sense of the word.
One must distinguish this invisibility from a mere question of access. It is not only that the tool be easy to reach; it is that once you use it, you cease to be aware that you are using it. It is a quality that does not reveal itself at first contact, but in repeated use: a tool can be easy to reach and stay visible, demanding each time that you think of it, pilot it, query it. The invisibility at issue here is deeper: it is the disappearance of the tool as an object of attention, to the point that you no longer distinguish its help from the natural course of work.
There is a paradox designers know well: the more successful an integration, the less it shows, and the less it shows, the less merit is attributed to it. The tool that effaces itself in the workflow never receives the compliments a flashy interface would have received. It is unfair, but it is also the sign that the work was well done. Invisibility is a success paid for in recognition: it serves the user without ever demanding their attention, and that is precisely why it lasts.
A successful integration receives no compliments. No one thanks what they forgot to notice.
Why this quality is never measured
If invisibility is the most reliable metric, one may ask why it appears in no evaluation. The reason is that it is, by nature, impossible to measure at the moment of evaluation. An evaluation happens at the start, over a few uses, in a setting where the tool is necessarily at the center of attention since one is testing it. Yet invisibility manifests only afterward, over time, when the tool has ceased to be an object of test and become part of the scenery. You cannot measure, by looking at it, a tool whose quality is precisely to make itself forgotten.
This impossibility has a perverse consequence: tools are selected on exactly the opposite criterion to the one that will decide their success. They are chosen on what they show, that is, on their visibility, when they will last or not according to their ability to disappear. The moment of selection values what the moment of use will punish. This is why so many tools impressive at purchase prove tiring in use: they were chosen for the quality that shows, and abandoned for lack of the quality that effaces itself.
Escaping this trap requires changing the question one puts to a tool before adopting it. Instead of seeking to be impressed, one must seek to project into repeated use: not “does this tool wow me now,” but “will I have forgotten it in six months while working better.” This projection is hard, because it demands imagining an experience the demonstration does not give. But it is the only one that predicts anything, because it interrogates the quality that truly matters at scale, and not the one that shines on the day of the trial.
The price of invisibility
This choice has a strategic consequence one must own. An AI designed to be invisible does not give a good five-minute demonstration; it does not lend itself to spectacular moments. It lends itself, by contrast, to a use that inscribes itself over time without weighing, and that is precisely what legal professionals ask for once they have tested enough tools to know what lasts and what does not. The spectacular seduces once; discretion serves a thousand times. And between seducing once and serving a thousand times, a serious organization knows what it prefers.
AI that signals itself ends up tiring. AI that becomes invisible ends up settling in.
This philosophy also has a consequence for how a tool is evaluated before purchase. If invisibility is the real criterion, then the five-minute demonstration is the worst moment to judge, because it rewards exactly the opposite: the spectacular, the visible, the demonstrative. A wise organization should not ask “what does it show?” but “what will it make me forget?”. How many fewer gestures, how many round trips avoided, how much attention returned to real work. These are questions no demonstration highlights, and that nonetheless decide everything.
One then measures how much the invisibility criterion overturns the way the market evaluates AI. Everything, in the way tools present and sell themselves, pushes toward the visible: one shows what impresses, one foregrounds what gets noticed, one optimizes for the demonstration moment. But what gets noticed in demonstration is exactly what will end up weighing in use. The market today rewards quality that shows, when the quality that lasts is the one that effaces itself. This gap between what is rewarded at purchase and what serves in use is one of the great confusions of the period, and it is the one that will resolve as organizations learn to distinguish brilliance from value.
It is this difference, barely visible at purchase and decisive in use, that structures what MAX builds, and it is in this direction that durable adoption lies. The market will eventually notice, because fatigue always ends up showing. A tool that signals itself endlessly wears out the patience it solicits; a tool that effaces itself leaves that patience intact, and it is this preserved patience that makes, in the end, the difference between a tool kept and a tool abandoned.
Do not ask what a tool shows you. Ask what it will make you forget.
