This piece first appeared in German. The English version is a rewrite rather than a line-by-line translation, and the German original stays online in the archive: KI verändert nicht primär Produktivität – sondern Organisationen.

A mid-term plan arrives at the committee already modelled: headcount requirements, investment, site questions, each with a scenario and a probability attached. Objecting to it now means arguing against something that adds up. The person with twenty years of context in the room finds that their judgment has kept its relevance and lost much of its force.

Artificial intelligence continues to be discussed along a familiar line: faster, cheaper, more efficient. That perspective is convenient because it connects to everything else, since it explains investments, justifies projects and reassures committees. It also misses the point.

The deep effect of AI sits where output is produced, and not in the output itself. It sits in the preparation of decisions: who decides on what basis a decision gets made, and how much contradiction still appears plausible. That is where organizations actually change, and AI works there as a structuring force and not as a tool. Quietly, systemically, permanently.

Plausibility makes dissent expensive

Organizations run on implicit rules, and anybody who has been there long enough knows whose judgment carries weight. Experience, overview, knowledge of context. Those authorities are rarely formal and thoroughly effective.

Deploying AI shifts that structure, because decisions increasingly get prepared on the basis of models: forecasts, scenarios, simulations, all of which look factual, rational and hard to attack. And the shift is exactly there. The more plausibly a decision has been prepared, the harder contradiction becomes, and not because the contradiction is wrong: it now aims at something that appears arithmetically coherent.

The decision itself stays human. Its shaping does not.

From judgment to probability

It used to be understood that decisions were made under uncertainty. Today uncertainty gets modelled, which is progress and an imposition at the same time.

Once probabilities dominate, the language of responsibility changes with them. Decisions stop being described as a weighing of options and start being described as a consequence of data, which relieves the decision-maker and moves something at the same time.

Mid-term planning shows it most clearly. Models deliver defensible recommendations for headcount, investment and location, they reduce variance and they produce consistency. What they do not deliver is context beyond their own assumptions. A company can still decide to hold capacity deliberately where the model suggests otherwise, and that choice now requires an explanation, rhetorically more than strategically.

Work does not disappear, it reorders

Debate about AI circles around activities: what falls away, what stays, what appears. That view is too narrow, because what matters is the work system a task sits inside, never the individual task alone.

In many knowledge-intensive organizations, AI systems now handle the preliminary work of analysis, research and structuring, which moves human work toward selection, evaluation and decision. Those activities are less clearly delimited than the ones they replaced, so roles blur and responsibilities become implicit.

The more preliminary work gets automated, the greater the responsibility of whoever decides, and the vaguer their formal role becomes. That vagueness is structural and not a transitional phenomenon, and it usually becomes visible only when organizations start reordering themselves.

Productivity is the acceptable story

When companies restructure or cut positions, AI often gets cited as the reason: efficiency gains, automation, changed competence profiles. None of that is wrong, and it is rarely the core.

In practice the questions underneath are different ones. How many decision levels does this still need? Which functions can be centralized? Where have interfaces grown up that coordinate more than they steer?

AI acts as a catalyst here, making visible what was already under organizational tension, and productivity is the accepted language for explaining those changes internally and externally. The actual rebuild concerns logics of control, never cost centers.

Reorganization follows the preparation of decisions

Many reorganizations happen after the introduction or scaling of AI systems, and seldom before. The reason is that those systems compress decision processes, where they rarely replace jobs directly.

Once planning becomes more central, faster and apparently more objective, decentralized structures lose legitimacy. Not necessarily their sense, and certainly their force, which changes the balance of power inside an organization.

Layoffs in that context look less like a technological necessity and more like the consequence of a new organizational logic. The technology supplies the justification, and the decision stays human while being framed differently.

Two objections worth taking seriously

Organizations have always been shaped by technology, runs the first one, and earlier waves of automation changed work without that being fundamentally problematic.

That is correct, and the difference lies in the depth of the intervention. AI does more than automate activities: it influences evaluation and decision processes themselves, and changes what counts as plausible, efficient or necessary.

The second objection holds that model-based decisions reduce arbitrariness and raise quality. Also correct, for as long as the models stay transparent and get understood as support. It becomes a problem where they turn into tacit authorities and displace responsibility in place of clarifying it.

Responsibility is relocated, never abolished

AI decides nothing; people do. They do it on a different basis.

Once models set the frame in which decisions appear plausible, the question of responsibility changes shape. It does not disappear, it becomes diffuse. Who contradicts a system that calculates consistently, and who carries the consequences of a decision that arithmetic seems to force?

Organizations that never negotiate those questions explicitly risk depersonalizing responsibility without actually making it objective.

What follows, without spelling it out

AI forces no particular decisions on an organization. What it forces is disclosure of that organization's own decision logic, which is uncomfortable and therefore frequently avoided.

Companies treating AI primarily as a productivity instrument tend to recognize its structural effect late, usually during a restructuring that then appears to have no alternative. Others take the opportunity to relocate responsibility deliberately and design where it sits. The difference lies in the relationship to their own organization, never in the technology.

So the real question may sit elsewhere. Less in how much work AI can take over, and more in which understanding of an organization it quietly presupposes, because where decisions increasingly arrive prepared, what still counts as decidable changes too.

This piece is part of a workshop series.

Sources

No external sources. Based on the authors professional experience.

The link has been copied!