The argument · open work · 2026
The Economics of the Other.
Eleven questions, each with a possible answer rather than a settled one.
Relational debt is a production liability, created when work can close without producing something the organisation will still need. Eleven questions, each with a possible answer rather than a settled one — so that if you disagree by the end, you know exactly where.
open studyThis argument needs other positions. Share a passage from your own work ↓I spent my first years in this work learning to predict people from the outside. I built models that grouped people who were similar enough that the same message would land on all of them. I could tell you who would leave that party with whom. What I never had access to was what the party meant to them.
Organisations are doing the same thing now, with better intentions. They measure engagement, sentiment, adoption. They measure AI: tokens, usage, hours saved. They know what their people are like, and how fast their AI is. They have no access to what the work is doing to them, and no category for what the work is failing to produce.
That split has always been there. Every organisation runs on two economies: one that makes the output, and one that makes what only people can make between them, which is what makes an organisation an organisation. The word comes from the Greek organon, an instrument, from the same root as ergon, work: parts that are different, arranged so that work can happen between them.
Working with teams as AI came into their work, I could feel it before I could name it. Then I could see it: the first economy getting faster, the second going quiet. And yet everyone is trying to fix it, with engagement surveys, culture programmes, collaboration training.
So I borrowed the logic of economics. Not because relationships are transactions, but because an economy is something we can work with: something we can see, count and value. If what forms between people has value, its absence has a cost. And if the work still needed it, something is owed.
These eleven questions are an attempt at that economy: the economics of the other.
1 · Why does every week go well and the year go badly?
A meeting booked by an agent. An agenda drafted by a model. Everyone arrives prepared, because everyone read a summary. A note-taker writes it up and sends the actions before the call has ended. Measured by anything an organisation measures, that hour was excellent.
A month later the decision reopens. Someone says but we agreed, and the agreement is in the summary rather than in the people. Elsewhere a junior still can’t handle a difficult client, and the manager who was never in that conversation finds out when the client is already angry.
Nothing went wrong in that hour. What went wrong is that the hour was supposed to make two things, and only one of them can be inspected afterwards.
2 · What were those people there to produce?
Not the decision — the summary records that. Not the document; it exists. Something else, which the hour was also meant to make: a judgement that had met another judgement, a commitment somebody took up in front of others, a standard that would still hold next week, a colleague who became a little more able.
Most of the time, I needed another person for two reasons at once.
Knowledge. Time. Skill. Information.
A judgement. A refusal. A commitment. A decision someone can stand behind.
By position I mean the place you occupy in relation to the work: what you are there to bring, decide, question and answer for. Legal’s position isn’t a person from Legal. It is the authority to say whether the contract holds, and to answer for that judgement if it doesn’t.
Take a junior who has to answer a difficult client. She goes to her manager. The manager asks a question, says no to one point, sends her back to rewrite. The answer gets better. And something else happens: the junior becomes a little more capable, and the manager now knows what is going on with that client. One trip did both jobs. Gregory Bateson defined information as “a difference which makes a difference.” The manager’s no was one.
Work produces an output, and it produces the capacity to keep producing outputs. Two economies, running at once. AI can increasingly supply the first reason I needed you. It cannot supply the second.
The task economy has a line in the accounts, an owner and a dashboard. The second — call it the economy of the other, because everything in it is made between positions — has never had a line anywhere. Not because nobody valued it, but because nobody could avoid it.
3 · So what changed when the answer started arriving in two minutes?
Not the value of that second thing. Its necessity. Before, a lack had nowhere else to go: I didn’t know, so I asked you; my argument was thin, so you pushed on it; I needed a decision, so somebody had to take it. The friction wasn’t designed. It forced positions into relation, and something formed there whether or not anyone intended it.
AI gives unfinishedness a second destination. What used to have to reach another position can now be completed without reaching it. The output is often excellent — which is exactly why this is hard to see.
4 · If the task still completes, what exactly was removed?
This is the question the whole method turns on, and it can be asked of any ordinary piece of work: what did this no longer have to pass through?
A proposal used to go past the account director before the client saw it. Now a model checks the tone and the numbers, and it goes straight out. What no longer passes through isn’t a check. It’s the one person who could have said “we can’t deliver this”, and who would have owned the promise if it went ahead.
A passage through another position was removed. The work no longer had to be exposed to someone who could question it, refuse it, alter it, or take responsibility for it.
Relational debt is what forms when the organisation still needs what that passage would have produced, and the work no longer produces it. That last condition is the guard: if the organisation no longer needs it, nothing is owed. Automate it.
5 · Then who is owed something?
Debt usually has a creditor: somebody waiting, somebody who will ask. Here, when the loop closes early, the person who would have been owed something simply never appears. The reviewer was never sent the draft. The colleague was never asked. Nobody is waiting, so nothing feels owed.
With agents it goes further. A study of 33,707 agent-authored pull requests found agents repeatedly abandoning the work once a human asked for a substantive change — the loop opened, and the thing that opened it could not be held to close it.
The debt has no visible creditor at the moment it forms. Nobody sends an invoice when a judgement wasn’t tested, a junior didn’t learn, or ownership wasn’t taken up. The creditor appears later: in the work that has to be reopened, the capability that has to be bought back, the manager who has to carry what nobody owned.
And where an agent is the author there may be no obvious debtor either: whoever reopens the loop is left holding something nobody experiences themselves as owing.
6 · If nothing breaks, why does it get worse rather than staying still?
A junior doesn’t merely miss one lesson. They learn how work is done here. A team doesn’t merely avoid one disagreement; it learns that polished closure beats unresolved difference. A manager doesn’t merely carry one unowned decision; the organisation learns that decisions can circulate without anyone taking them up.
The debt is what wasn’t produced. The interest is what forms in its place. And when that interest accumulates, it begins to look like culture.
7 · Isn’t that what a human in the loop is for?
The phrase comes from control engineering: an operator kept inside an automated system to stop it if it goes wrong. It puts a person at the end, as a checkpoint.
But approving is not producing. The manager who signs off the model’s answer is not the manager who said no, sent it back, and learned what was happening with that client.
Or take Legal’s view. Ask the AI, and in two minutes you have a clear, useful answer. Nobody waited. Nobody disagreed. But Legal never saw it. Nobody there will stand behind it, and nobody there knows what you are planning.
The work doesn’t need a human in the loop. It needs the other side in the work.
A human can inspect an output without becoming implicated in its production. So the question is not whether a human touched the loop. It is whether another position could make a difference to what was produced — and whether that difference had consequences.
8 · When does another person actually change anything?
Not by being present. Not by being consulted. Between a lack and what gets produced there is a chain, and most remedies fail because they aim at the wrong link.
- lackSomething can’t be closed from this position.
- desireThe lack acquires a direction: there is something I want from you.
- reachThe distance is crossed — or made crossable by how the work is set up.
- passageWhat I bring stops being under my control. Another position can answer, contest, refuse, change it.
- productionSomething exists that neither position could have made alone.
A planner isn’t sure a campaign promise can be kept. lack She wants to know from the service lead. desire Thursday’s review has a slot where service sees drafts before they go. reach The service lead says no to one line, and the planner rewrites it. passage What goes out is a promise service can keep, and both of them know why. production
Passage is where implication begins. I have not merely reached another person: I have put something of my own position into relation with theirs, where their response can matter to what I do next. That is what I mean by implicarsi — not another link in the chain, but the quality of this one.
Contact is not passage. Presence is not passage. A passage happens when neither what was brought nor the position that brought it can stay entirely unchanged. Nothing has to change every time — but a no has to be possible, and it has to have consequences.
a role counts only if its refusal could have changed what happened next
9 · Whose job is any of this — the person’s, or the company’s?
People are told to collaborate more, to ask for help, to speak up. Then the same people are given a deadline that makes asking expensive, a meeting they can attend but not shape, and a metric that counts what they shipped alone.
A team is told to speak up, then given a two-day turnaround on a pitch. Asking the strategy lead means waiting until Monday; the model answers now. Nobody decided the strategy lead doesn’t matter. The deadline did.
People do the reaching. Organisations determine much of what reaching costs. Time, mandate, legitimacy, workflow and consequence can make another position necessary, optional, or practically unreachable. That is the part that can be designed.
10 · Can you design desire?
No — and the attempt produces the version that hides. But desire is sustained rather than ordered: it survives wherever a lack is still treated as something that matters. A gap allowed to stay open becomes a question addressed to someone. A gap closed the moment it appears never becomes a reason to reach for anyone.
In one version of a weekly review, each person opens with a doubt instead of an update: one thing they don’t know yet. It takes ten minutes. It’s where the questions that used to go to the model at six in the evening start arriving.
Three things can be designed. Protect the lack: can something stay unfinished long enough for the work to discover whether another position is actually required? Redesign the reach: is the other worth going to, in time, mandate and legitimacy? Redesign the passage: once reached, can their difference change what gets produced? In practice that is three moves with a team: sit with where the passage isn’t there, weigh what it costs, and bring the other part back in where it still makes the difference.
11 · How would you know if I’m wrong?
Four ways, and they are not rhetorical.
- If organisations need much less of this than I assume. Then most of what I call debt is a cost correctly removed, and the practice has to be able to say so in the room.
- If the capacity forms anyway, somewhere else. Simulation, volume of attempts, feedback from models. Then the passage mattered less than the formation, and I have confused a mechanism with a habit.
- If the bill never arrives. If work that systematically removes these passages shows no later loss of judgement, ownership, formation, standards, coordination or capability — and no equivalent cost appears anywhere else — then what I am calling relational debt is not debt.
- If the designed passage doesn’t produce what the accidental one did. This is the risk inside my own remedy: friction that was structural may not survive being made deliberate. Then the diagnosis stands and the intervention doesn’t.
The strongest counter-evidence available today is a field experiment at Procter & Gamble, where individuals working with AI matched the performance of teams working without it (Dell’Acqua et al., Organization Science, 2026). My answer isn’t to deny it: one position plus AI can now produce what used to take two — for work one position can produce.
Five questions for any passage AI now touches
- 1What is the luxury here?
- 2What is the cost?
- 3Who enjoys the luxury, and who pays the cost?
- 4Can we afford it?
- 5And for how long?
And the question I would leave you with
Not what humans should keep doing. Not where to keep one in the loop. Something you can ask about one passage of work, this week:
What does this still need another role to produce — and not just to finish?
If the answer is nothing, automate it. Take the cost out.
If the answer is something, don’t preserve the old friction. Design the passage that still needs to exist.
That is the work of altraparte.