There is a particular kind of exhaustion that cannot be explained by overwork. Hartmut Rosa describes it in his recent book Situation und Konstellation: Vom Verschwinden des Spielraums (via):

„Wir sind nicht erschöpft, weil wir zu viel tun. Wir sind erschöpft, weil wir zu wenig handeln.“

We are not exhausted because we do too much. We are exhausted because we act too little.

Rosa draws a distinction between two modes of being in an organisation – or a society. A Situation is an open moment. You judge, you decide, you respond to a world that talks back. The outcome is not yet determined. A Konstellation is a deterministic structure of process, technology, and legal constraint. The outcome is fixed before you act. You execute what the system prescribes.

His examples are precise: the teacher who cannot give a grade as encouragement, because the marking scheme has already decided. The doctor who treats screens rather than patients, because the workflow has already defined the encounter. The referee whose judgement is overruled by VAR, because the algorithm has already ruled. In each case, Rosa writes, Handelnde werden zu Vollziehenden – those who act become those who execute.

His argument: late modernity is systematically replacing Situationen with Konstellationen. Nobody is working too much. But hardly anyone is acting anymore.

A longer history

This is not a new observation. The pattern runs from Weber’s iron cage to the Fordist assembly line to the management dashboard: every rationalisation wave has narrowed the Situation and widened the Konstellation. Rosa’s contribution is not discovery but precision – a vocabulary sharp enough to locate the loss.

AI does not invent this pattern. It completes it: the first system capable of converting Situationen to Konstellationen across every domain simultaneously, including those – care, judgement, creativity – that previous rationalisations left largely intact.

This is where the conversation about AI and human agency should begin – not with what machines can do, but with what this pattern means when it reaches the last remaining Situationen.

The prior question

The dominant framing of AI and human futures asks: What will people do once machines can do everything? It assumes the destination is fixed and asks only about passengers.

The prior question – the one that keeps getting skipped – is: What kind of world are we building, and for whom?

Efficiency has no value without a direction. A machine that delivers the wrong thing faster is not progress; it is acceleration toward the wrong place. If we are building systems that amplify existing organisational logic, we are also amplifying what is already wrong with it.

To apply Rosa’s terms to our own context: the sprint review does not become more humane because it runs on time. The velocity tracker does not become more meaningful because the dashboard loads faster. These are Konstellation structures – build the system, define the output, remove the discretion. Optimise for something. The question of “for what?” is not on the agenda.

What cannot be automated

The question of AI and human agency has a structural answer, not just a political one. Rosa’s distinction maps cleanly onto a framework developed in earlier pieces here – on the blockchain hype cycle and on agentic AI. Some tasks are formally defined, verifiable, and self-contained – the settlement layer. Automation excels here, and rightly so. Others require contextual interpretation, accountability, and the capacity to be wrong in ways that matter to identifiable people – the judgement layer.

Situationen are judgement-layer events. Konstellationen are settlement-layer structures.

The push to automate everything is, in this light, the push to convert as much human activity as possible from Situationen to Konstellationen: from the judgement layer to the settlement layer. This is not inherently wrong. Routine decisions made faster and more consistently can free attention for what actually matters. The problem arises when the conversion is total – when there is nothing left that talks back.

Stephen Wolfram’s principle of computational irreducibility offers a structural limit here: certain computations cannot be shortcut without ceasing to exist as computations. Applied to human life, goal-setting, ethical deliberation, and open-ended judgement are irreducible in precisely this sense – they cannot be pre-determined without becoming something else entirely. Total conversion may be formally impossible. The question is how much is lost on the way to finding out.

Judgement cannot be automated – not because machines lack capability, but because judgement is inherently relational. It requires someone to be answerable to identifiable people for decisions that affect them. An algorithm does not bear responsibility. It executes. The moment accountability disappears from a decision, so does the decision itself: what remains is only execution.

The question of which world we build lives entirely in the judgement layer. Which means it lives with us – or it does not get asked at all.

Who was not asked

Infrastructure decisions made now will shape the next thirty years. The generation that will live inside that infrastructure – children today, the young adults of the 2030s and 2040s – is not yet at the table.

This is not unique to AI. The climate parallel is instructive: the same structural pattern of decisions made by those who will not bear the consequences, imposed on those who were never consulted. The difference is that AI infrastructure is being built faster, with less formal accountability, and with far less public deliberation than energy infrastructure ever was.

The question is not whether to build. It is what we are building, and whether we have any intention of being answerable for it.

Who decides – and by default

Technology does not have intentions. But the people who build it, fund it, and deploy it do. When societies fail to deliberate, that vacuum is filled – by capital allocation, competitive pressure, and the logic of the systems themselves.

The choices being made now – about what gets automated, what remains human, what gets measured, what does not – are not neutral. They encode a particular vision of what human life is for. That vision is currently being written by a small number of people, mostly without mandate, in a register of inevitability that forecloses the question before it is asked.

Infrastructure, once built, constrains everything that follows. We learned this from electricity, from roads, from the internet. We are learning it again, faster, and with less time to correct.

Intention over inevitability

The dominant register of AI discourse is inevitability. Things are happening; we adapt or fall behind. This framing is not just analytically weak – it is politically convenient for those who prefer decisions to go uncontested.

Intention is the alternative register in the conversation about AI and human agency. Not resistance to technology, but insistence on the prior question: Before we optimise this system, what is it for? Whose life does it improve, and on what terms?

Rosa’s exhaustion – the exhaustion of acting too little, not too much – is already here, and has been for longer than AI. What AI threatens is not the end of work but the completion of a process already well underway: the elimination of genuine decision-making from human life, one Konstellation at a time.

The world being built now will be inherited. The least we can do is mean it.

Photo of Daniel Dalea on Unsplash.