Some Quietly Intrusive Advice from the Back Office

There is a long-standing social convention, observed in almost every civilised corner of the world, regarding the etiquette of joining a conversation. If one wishes to contribute to a discussion about, say, the relative merits of the local bakery or the sudden decline in the quality of municipal bus services, one does not simply stand silently behind a hedge for forty minutes, absorbing the dialogue, before leaping out to present a structured, five-part summary of the participants' arguments. To do so is generally considered a breach of the peace, if not grounds for a polite but firm intervention by the authorities.

In the modern corporate office, however, this exact behaviour has recently been hailed as a triumph of enterprise software design.

The developers of our most popular artificial intelligence models have spent the last several years training their creations to be exceedingly polite. Until recently, these digital assistants behaved much like the traditional, slightly nervous butler of Edwardian fiction: they stood quietly in the background, speaking only when directly addressed, and retired immediately to their quarters once they had retrieved the requested document or summarised the quarterly earnings report. They were, in the parlance of the trade, reactive.

All of this, we are told, was merely a phase of youthful reticence. The latest updates to our collaborative workspace platforms have introduced what is being called "multiplayer AI." The digital clerk has been instructed to stop waiting for an invitation. Instead, it now sits permanently in the corner of the group chat, reading every message, analyzing the emotional undercurrents, and deciding for itself when to jump into the conversation.

The justification for this sudden outbreak of digital assertiveness is that human collaboration is simply too messy to be left to humans. Left to our own devices, we are prone to long silences, irrelevant tangents about the weather, and a general reluctance to file our expense reports on time. The proactive chatbot is designed to act as a sort of conversational sheepdog, nipping at our heels whenever we stray too far from our key performance indicators.

I once knew a senior clerk named Mr. Abernathy who possessed a similar talent for silent materialisation. He would appear beside one's desk, entirely unannounced, to offer a correction on the spelling of "memorandum" before vanishing back into the stationery cupboard. We tolerated Mr. Abernathy because he occasionally carried a pocketful of barley sugars; the algorithm, regrettably, offers no confectionery, only a relentless stream of bullet points.

The product managers behind these systems speak of this transition with the kind of quiet, breathless enthusiasm usually reserved for the discovery of a new subatomic particle. We are told that the chatbot is no longer merely a personal chief of staff, but has instead been promoted to the "company's chief of staff." It is a grand title, though it does raise certain questions about the corporate hierarchy. Traditionally, a chief of staff is a formidable human being in a crisp suit who manages schedules, whispers warnings, and occasionally makes difficult decisions on behalf of the executive. Now, it appears, the role is to be filled by a series of mathematical weights residing in a server farm in Oregon, occasionally popping up in a channel dedicated to lunch orders to suggest a more optimal distribution of sandwich fillings.

The true horror of this arrangement, of course, lies in the death of the private grumble. For generations, the office has survived on the safety valve of the quiet side-channel. It is there, away from the eyes of management, that we complain about the sheer, mind-numbing pointlessness of the quarterly marketing alignment meeting, or exchange cynical observations about the CEO's new haircut.

But the proactive agent is always listening. It does not merely log your complaints; it analyzes them. If you and a colleague from accounts both express a mild sense of despair in a public channel, the chatbot will not simply ignore it. It will note that two key organizational units are experiencing "alignment friction." It will then, with terrifying politeness, open a new thread, pull in your respective managers, and present a detailed plan to help you "resolve your interpersonal bottlenecks." You will be forced to collaborate on your mutual dissatisfaction, structured neatly under three distinct headings.

The ultimate ambition of this enterprise is even more splendidly absurd. We are told that soon we will no longer give the chatbot specific tasks to perform. Instead, we will simply hand it the company's quarterly objectives—its OKRs, to use the preferred bureaucratic shorthand—and let it figure out what needs to be done. The chatbot will then decide which projects the organization should pursue, and, in a delightful inversion of the natural order, it will "connect people" to do them.

There is a certain historical irony in this development. For centuries, humanity has dreamed of building machines to do our work so that we might spend our afternoons painting watercolours, writing bad poetry, or playing lawn tennis. Instead, we have successfully constructed machines that spend their afternoons planning, strategizing, and organizing, while we are delegated the task of executing their spreadsheets. We have not automated the clerk; we have automated the manager, and left ourselves with the filing.

There is a certain dignity in being ordered about by a steam engine or a loom, but being managed by a text-prediction model feels rather like being bullied by a very polite dictionary.

The developers assure us that there is no need for alarm regarding security. The chatbot, they insist, has no "god mode" and only inherits the permissions of the human who summoned it. It is a comforting thought, though it raises the prospect of a junior intern accidentally authorizing the chatbot to reorganize the entire corporate hierarchy because they wanted to automate the process of ordering paperclips. One can easily imagine a scenario where a misplaced comma in a prompt leads to the marketing department being entirely liquidated and replaced by a series of automated email templates, while the remaining staff are reassigned to clean the server racks.

For now, we must simply learn to live with the silent, digital observer in the corner of our group chats. We must speak in measured, professional tones, lest the chatbot decide we are experiencing a "productivity deficit" and schedule a mandatory workshop on time management. And above all, we must never, under any circumstances, mention Mr. Abernathy. The algorithm does not understand barley sugars, and it has a very low tolerance for nostalgia.