What the Assistant Makes Visible — Four Tiers of Reciprocity
Tenth text in the cognition / doctrine series. After the end of the junior consultant, a question that arises at the individual scale: why have so many executives who tried the assistant closed the tool saying it was useless, and why do those who extract a decisive return from it not always know how to explain what they put into it.
The Threshold of Reciprocity
Everywhere you read that the assistant saves time. The figure circulates — sometimes thirty percent, sometimes forty — never backed by anything beyond a vendor survey of its own users. And yet, around us, in executive committees as in project leadership, the verdict is quieter and less flattering: most of those who tried it dropped out. They opened the tool, called on it a dozen times, received answers half of which smelled of the factory and half of which saved them a few minutes at the cost of a painful review, and closed it telling themselves they would try again later, when the product was better. This explanation is convenient. It is also false.
What was at stake in the abandonment was not the quality of the model, already superior to what most will admit. It was an asymmetry of expectation. The user believed they were dealing with a service; they were dealing with a relationship. A service gives back at once what you ask of it, against payment. A relationship presupposes that you first gave it something, against nothing. The assistant belongs to the second register far more than to the first, and it is there, in that initial confusion, that everything is decided. No one trains a tutor by first asking for a demonstration of competence; you begin by laying out what you expect, what you already know, what you are looking for. The effort is one-sided for weeks before becoming mutual, then profitable.
What we want to hold here is a trajectory in four movements, whose final return is less the efficiency gained than the lucidity acquired. The assistant, at the end of this trajectory, does not merely make us go faster. It makes us see. But what it renders visible appears only at the fourth tier, and the majority of users stop at the first.
Give Back to It
The first movement consists in giving before asking. This seems counterintuitive for someone paying a monthly subscription; it stops being so the moment you accept that what costs you, in the use of an assistant, is never the token consumed but the context absent. A generalist model without context produces an average. The same model, instructed in our role, our constraints, our vocabulary, our archives, produces an answer that resembles what we would have formulated ourselves had we had three hours. The whole difference lies in the initial investment.
This investment has a name in recent Anglo-Saxon literature. People speak of a memory layer, of personal knowledge management, of a second brain. The terms float, and everyone dresses them as they please. What they designate is invariant: an externalized layer where our tacit knowledge lives — that is to say, what we know without formulating it, what we do without describing it, what constitutes our craft and which no one, ever, has asked us to write down. Polanyi called this tacit knowledge and held that it was irreducibly personal. The assistant's proposition is less ambitious and more unsettling: that knowledge can be made operable by a third party, on the condition that it has been made explicit.
The operation is not spectacular. It consists in noting, in filing, in tagging, in dating. It consists in writing a file that says who we are, what we are working on, which decisions have been made, which trade-offs have been settled. It consists in maintaining a trace of past conversations and project contexts. Nothing glorious. This work is even thankless, for it produces nothing visible for weeks, save a folder that grows fat and whose return remains hypothetical.
And yet this work is the condition of everything else. Without it, the assistant is no more than a sophisticated spell-checker. With it, it becomes an interlocutor that inherits, at every session, what we have built. The difference between the two is not one of degree but of nature. It explains why certain executives draw from the tool a return that seems implausible to the others, even though they use the same model, the same version, the same subscription. The former gave something back. The latter were waiting for a service.
Work With It
Once the context layer is minimally constituted, comes the second movement, and it is probably the most thankless. You formulate tasks for the assistant, you look at its outputs, you correct, you rephrase, you start again. This phase resembles that of a manager training a brilliant but inexperienced junior: everything has to be said twice, you must explain what is self-evident to us, dismantle our own habits in order to make them explicit.
Vygotsky, who conceived of learning as a zone of asymmetry between what a child can do alone and what they can do with a scaffolding adult, would have recognized here an analogous configuration, inverted only by the fact that it is the human who scaffolds the machine. We lend the assistant our judgment, our eye for what is fitting, our sense of the situation. In return, it forces us to say aloud what we were doing in silence. What seemed obvious ceases to be so the moment it must be formulated. What we took for an intuition reveals itself, under the labor of explicitation, to be a rule or a heuristic we can teach.
This phase lasts, and it is here that the majority abandons a second time. The cost-benefit ratio looks unfavorable: we spend more time correcting than we would doing it ourselves. The calculation is right in the short term and wrong in the medium term. For what is built during this phase is not an immediate productivity gain. It is a reciprocal understanding. The assistant learns to recognize us — in our turns of phrase, in our preferences, in what we validate and what we refuse. We learn to call on it where it will be good, to avoid the zones where it will be soft, to rephrase our requests so that they are operable.
Sennett, in his book on craftsmanship, describes the long patience by which an artisan acquires their gesture. He insists that this gesture is not learned through principles but through the repetition of work done jointly with a more experienced companion. The phase of co-construction belongs to the same register. We train the tool and we train ourselves to the tool, simultaneously, without being able to distinguish the two operations clearly. And on emerging from this phase, something has changed that no tutorial could have conveyed to us: we know what to entrust to it, and how.
Let It Work With Us
The third movement is the one in which the assistant begins to take charge of entire sequences, but under our gaze. Reviewing the meeting minutes, drafting the delicate email, summarizing the long thread, preparing the meeting brief, the internal note that runs to three pages: so many tasks we would have entrusted to a human only at the cost of a coordination time greater than the time to do them, and which we here let fall into a framed delegation.
It is in this phase that the first quantifiable gains appear. Thirty minutes regained on preparing a steering committee, an hour on the technical brief for a consultation, forty-five minutes on the framing note destined for the committee. These gains are real, and if you began to add them up, you would reconstitute a part of the day. But this is not the most interesting thing. The most interesting lies in what we discover along the way and did not expect.
We discover, for instance, that the assistant is better than us at certain tasks we believed were ours. Summarizing a long text, reformulating an argument, producing a first structured version: these operations, in which we had developed a competence we were quietly proud of, are accomplished by the tool with an average quality superior to our own. The observation is unpleasant. It forces us to ask anew the question of our added value, and to recognize that it lodges elsewhere than we thought.
We also discover, conversely, zones where the assistant is frankly weak and where we remain indispensable. Grasping the unsaid in a client message. Arbitrating between two options whose political consequences are measured against a history no one has written. Detecting the moment when one must stop debating and decide. These competencies do not migrate. They define, in negative, what remains irreducibly within our purview, and which no foreseeable improvement of the model will come to absorb.
This double discovery — these zones where we are more dispensable than we thought and those where we are less so — silently restructures the representation we held of our craft. It is already a first audit. But it is still only a preamble to the one that comes at the next tier.
Let It Work Without Us
The fourth movement is by far the hardest to cross, because it touches the psychological relationship we maintain with our own activity. It consists in entrusting the assistant with entire sequences that we no longer supervise in real time and of which we receive only the result. Planning a week. Following an inbox. Preparing the team check-ins. Sector intelligence. The weekly summary of project activity. So many tasks we used to do without even naming them, because they had sedimented into our routine to the point of becoming invisible to ourselves.
Entrusting these tasks is not trivial. It amounts to acknowledging that they did not require our intervention. And that acknowledgment, which for a consultant or an executive is a deep revision of the representation of their daily usefulness, does not happen in a day.
Illich, in his essay on convivial tools, theorized the existence of a threshold beyond which a tool ceases to serve man and begins to dispossess him. His critique targeted the school and medicine. It can be turned around here, and this is what makes the moment delicate: as we let the assistant work without us, we are not dispossessed of our work, we are dispossessed of the illusion that it had to be us who did it. A subtle distinction, an unsettling consequence.
For what we recover in liberated hours, at this last tier, is immediately absorbed by the question thrown back at us: what to do with this time now available? And, more profoundly still: how much time, then, had we spent up to now on tasks we believed required our judgment and which required only the appearance of it? The stupor is brief, and the need to mask it sometimes strong. Many, at this stage, take back in hand tasks they had no reason whatsoever to take back, simply to reconstitute the feeling of a full schedule.
Coda — What One Sees Then
At the end of these four tiers, what is revealed to us is not what the assistant can do. That, we had guessed from the start, and the industry has never ceased drumming it into us. What is revealed is what we were doing, and were not seeing. The audit did not bear on the tool. It bore, by ricochet, on the use we were making of our days.
And the verdict is uncomfortable. A significant part of an executive's work, as we exercised it, was an activity of filler that had never been looked at. Not useless, strictly speaking: useful in the sense that it sustained relationships, marked presences, gave signs of organizational vitality. But useful without judgment, without arbitration, without the share of discernment we believed we put into it. The assistant does no better than us on that part, because that part asked no one to do better; it asked only that it be done.
Drucker, who had coined the expression knowledge worker in the nineteen-fifties, saw in intellectual work an unprecedented challenge: that of an activity whose productivity could no longer be measured by the instruments of industrial production. He sensed that the principal difficulty would be to distinguish, within the continuous flow of the days, what arose from judgment and what merely produced work. He did not, in his time, have the instrument that would make this distinction operable. We have it.
The instrument does not tell us what we must do. It tells us, by subtraction, what we no longer needed to do. And it is in that hollowed-out space that the question lodges which had remained, until then, masked by the background noise of daily activity: what are we good for, really, once the appearance of being busy has been removed. The question is new for most of us. It is also more useful, for thinking about the craft of the next ten years, than the saturated question of whether AI will replace us. It silently rephrases it: what was it that deserved that we do it.
It is only at that moment, and for those who will have crossed the four tiers, that the assistant will have kept its promise. Not the banal one of having saved time. The other, rarer one: of having made visible what was worth our time.
TITAN, TenderGraph's cognitive system, applies these best practices — language science, rhetoric, cognitive science — to every bid it works on. And TenderGraph helps teams optimise their pre-sales process along these lines: discover TenderGraph · talk to our team.
Bibliography
On tacit knowledge and explicitation
- Michael Polanyi, The Tacit Dimension, 1966.
- Lev Vygotsky, Mind in Society, 1978 (on the zone of proximal development).
On craftsmanship and long apprenticeship
- Richard Sennett, The Craftsman, 2008.
On the knowledge worker and intellectual productivity
- Peter Drucker, Landmarks of Tomorrow, 1959 (first use of knowledge worker).
- Peter Drucker, Management Challenges for the 21st Century, 1999 (chapter on knowledge worker productivity).
On tools and their threshold of reversal
- Ivan Illich, Tools for Conviviality, 1973.