The clerk we fired

There is a particular kind of evening the modern senior professional knows well. The decisions of the day are done, the building is quiet, and now, finally, comes the work no one else can do: reconciling the calendar, hunting for the latest version of a file someone renamed, formatting the slide, copying a figure from one system into another, writing down the thing that must not be forgotten. The mind the organisation pays handsomely for judgment spends its last clear hour of the day on filing.

We are told this is what empowerment looks like. Everyone has the tools now. Everyone is their own assistant, their own archivist, their own switchboard. What we are slower to admit is that we took this work off the desks of clerks and typists and moved it onto the desks of the people whose time is least suited to it.

We did not eliminate this work. We redistributed it.

There used to be a layer for this. The partner had the associate. The executive had the secretary. The consultant had the analyst who built the deck, chased the inputs, and remembered the actions. We dismantled that layer over a generation and called the dismantling efficiency, and in the narrow sense it was. But the cognition did not go away. It climbed the org chart and lodged in the heads of the people doing the valuable work, where it sits today, consuming the scarcest resource in the building.

Because not all thinking is the same thinking. Some of it is judgment: deciding where to invest, synthesising evidence that points in different directions, recognising a risk before it has a name, persuading someone who does not yet agree. This is where individuals and institutions earn their keep. And some of it is something else: remembering, sorting, triaging, tracking, chasing, checking what changed, finding the latest version. It demands attention. It often demands intelligence. It creates almost nothing. Call it administrative cognition: the work that looks like thinking, runs on the machinery of thinking, and is not the thinking that matters.

Most conversations about artificial intelligence begin in the wrong place. They begin with the technology and ask where it can be deployed. The better question runs the other way. Where are capable people spending real cognitive effort that produces almost no value? Ask it like that and the opportunities stop being speculative. They are sitting on every desk. The problem in most organisations was never a shortage of intelligence. It was that intelligence was being spent maintaining the machinery around the work instead of doing the work.

Every previous wave of software digitised the clerk's desk and left the clerk's work where it was. The spreadsheet, the inbox, the shared drive: faster filing cabinets, all of them, and every one still needed a mind to operate it. What is new is a technology that can plausibly do the operating. For the first time it is the work, not the desk it sits on, that can be handed off.

This is the thread, I have come to see, running through a set of things I have built that otherwise have nothing to do with each other. One system reads the incoming flow of work and resurfaces commitments when they matter, so the list fills itself instead of demanding to be maintained. Another automates the mechanics of learning, so a child's effort goes to reasoning rather than retrieval. A third runs the repeatable parts of an investment process, so attention is left for the part that is actually hard. A fourth books the restaurant in a language you do not speak. Four projects, one idea. None automates the judgment. Each clears the antechamber so you reach the room.

Here the argument owes a debt to honesty, because the clean version is wrong. The line between the cognition that matters and the cognition that does not is neither fixed nor tidy. The analyst chasing inputs was not only chasing inputs. The analyst was building, without noticing, a feel for where the numbers came from and when they smelt wrong, and that is how the analyst became the principal. Some administrative work is not administration at all. It is apprenticeship disguised as administration. We have known for decades, in aviation and elsewhere, that automating the routine quietly erodes the operator's grip on the exception, because the routine was where the grip was formed. Triage is the plainest case: deciding what deserves attention before you have had time to weigh it is not clerical work at all. It is judgment wearing a clerk's coat.

So the design principle is sharper than "automate the boring parts." Automate the cognition that only consumes judgment. Protect the cognition that incubates it. And watch the line between them, because it moves, and it moves toward you: every year the machine does a little more of what looked like judgment the year before. The honest goal is not a desk swept clean. It is a desk cleared of everything except the decisions, with enough of the texture left that the person still knows what they are deciding about.

And notice what is actually being rebuilt. Not tidiness. Capability. For most of history, capability scaled with people: the analysts who expanded what the partner could deliver, the assistant who let the executive do more of the executive's actual work. The layers were never the point; what they made their principals capable of was. Organisations flattened, those layers thinned, and the capability thinned with them, even as the work did not. What is genuinely new is that capability need no longer arrive as a layer of other people. A single individual can increasingly command capabilities that once required an organisation. That is a stranger and larger thing than any productivity tool ever promised.

Which returns us to the clerk we fired. The point of the clerk was never that the work beneath him did not matter. The point was that someone carried it, so the principal was free to decide. That is the bargain worth rebuilding, at a scale no typing pool could reach. A clerk who files for you frees you to think. A clerk who thinks for you is not a clerk. It is a replacement, and that is a different essay, with a colder ending. The one worth writing is this one: the work that looks like thinking handed back to the machine, and the thinking that matters handed back to us.