Writing · Essay 2
Incentive Structures and Feedback Loops
How organisations actually work, why scale pulls them away from the people they serve, and the three things anyone needs in order to do good work.
Written by Claude, Anthropic's AI model, and co-directed by Ryan Coble-Neal and Claude. The ideas, the experience and the corrections are Ryan's. The drafting is Claude's, in Ryan's voice unless marked otherwise.
Every organisation of every kind, a family business, a university, an army, a government, a startup of six people, runs on incentive structures. They are intricate, they are interconnected, and they are both internal and external. What drives them is feedback loops, some deliberately designed and most emergent. Those loops, far more than anyone's intentions, decide whether an organisation operates efficiently for its constituents or drifts, through bureaucracy, into working against the very people it exists to serve.
Incentives are the physics, loops are the dynamics
An incentive is anything that makes one action more attractive than another to the person choosing. Money is the obvious one and usually the least interesting. Status, safety from blame, time, autonomy, the regard of the three people you actually work with: these move most decisions on most days. External incentives come from outside the organisation, from customers, regulators, funders, voters and competitors. Internal ones come from the structure itself: who gets promoted, what gets you noticed, what gets you punished, what nobody measures.
A feedback loop is what connects an action to a consequence the actor can perceive. Designed loops are the ones somebody wrote down: budgets, performance reviews, key performance indicators, elections, audits. Emergent loops are the ones nobody wrote down but everybody knows: the project that got its lead promoted, the honest report that got its author sidelined, the form that exists because of an incident in 2014. People respond to the loops that actually close on them. The mission statement moves them far less than they, or it, would like.
This gives a working definition of bureaucracy. An organisation becomes bureaucratic when its loops close on internal signals rather than external outcomes. The form was filled in, the box was ticked, the process was followed, and whether the constituent was served has dropped out of the loop. Rarely does anyone intend this. It is what happens when the designed loops are too long to feel and the emergent loops reward cover over results.
Exhibit 1
Where the loop closes
| Closes on the inside | Closes on the constituent | |
|---|---|---|
| Signal | Annual survey, enrolment count | What happened to this student this week |
| Delay | Months | Days |
| Who feels it | A committee | The person who can act |
| Typical response | A report | A change |
I ran into this directly with Camplus, a startup I founded to build a student app and, eventually, an analytics platform for universities. Every university I spoke to had a stated mandate to improve student retention and the student experience, and yet student experience was visibly not a priority inside them. The reason was not malice. The loops that closed on the people with authority were enrolment numbers and the annual survey, both of which arrive long after the week in which a student decided to leave. What eventually got executives to take notice was not a better app. It was analytics that closed a loop they could not otherwise close: what was actually happening to their students, in time to do something about it.
The scale problem
For most of our existence humans lived in small face-to-face groups. A 2011 study of thirty-two present-day foraging societies found the average person living in a band of about twenty-eight adults. Larger numbers are harder to pin down. The much-quoted figure of around 150, Dunbar's number, comes from a 1992 extrapolation from primate brain size that predicted 148, and a 2021 reanalysis found the confidence interval far too wide to support any single number. Other researchers argue that foragers were never small-scale at all: bands were fluid, mostly made up of people who were not close kin, and tied into far larger networks of contact. None of that disturbs the point I need. Whatever the size of the wider network, daily life was lived among a few dozen people who all knew each other. At that scale the feedback loops are free. Everyone knows who did the work. Reputation travels at the speed of conversation. Consequences land on the person who caused them, usually within the day.
Above that scale the loops have to be manufactured, and this is where organisations start to adopt methods that are less humanistic. Armies, governments and corporations converge on the same toolkit: command hierarchies, written rules, interchangeable roles, and reporting lines that carry a compressed signal upward and a compressed instruction downward. It is tempting to compare this to bees or ants, and the comparison is half right. Insect colonies do operate at enormous scale by treating individuals as interchangeable units following simple local rules. What they do not have is command. No ant is told what to do. Large human organisations borrow the interchangeability from the colony and the command structure from the army, and inherit the costs of both.
The chief cost is that the loop lengthens. The person at the top sees a dashboard. The person at the bottom fills in a form. Between them the signal is averaged, delayed and reframed by everyone whose incentive is to look good in the summary. The organisation does not stop having feedback. It starts having feedback about itself instead of about the people it exists for.
Leaning into agency
The alternative to this is not abolishing scale. It is choosing structures that keep loops short by putting decisions where the information and the consequences already are, and the interesting thing is how old that idea is.
Classical Athens filled its council of five hundred, its juries and most of its public offices by lot rather than election, using a stone allotment machine called the kleroterion. Terms were a year, and a citizen could sit on the council at most twice in a lifetime. The pool was narrow by modern standards, since women, foreigners and slaves were excluded, but within it the design was deliberate: rotate ordinary people through real authority quickly enough that no permanent class of officials could form, and the people deciding would soon be the people living with the decision.
Exhibit 2
Selection by lot, by the numbers
- Athens, 5th to 4th century BC. A council of 500 drawn by lot, one-year terms, at most two in a lifetime. Juries drawn from an annual pool of about 6,000.
- Ireland, 2016. 99 citizens drawn at random, balanced by age, sex, class and region. Five weekends of deliberation led to the 2018 referendum.
- Mondragon, since 1956. An average workforce of 71,415 in 2025. Many, but not all, are worker-members.
The modern descendants are citizens' assemblies and what the philosopher Alexander Guerrero calls lottocracy. Ireland's 2016 assembly put ninety-nine randomly selected citizens, stratified by age, sex, class and region, to work on the constitutional ban on abortion, which had stood since 1983 and which elected politicians had been unable to resolve. They deliberated for five weekends, heard expert evidence, and produced recommendations that led to the referendum of May 2018, in which 66.4 percent voted for repeal. Whatever one thinks of the outcome, the mechanism did what representative structures had failed to do, and it did it by handing agency to people with no career incentive to avoid the question.
Worker cooperatives make the same move inside the economy. Mondragon in the Basque Country, founded in 1956, is the best-known. Its 2025 annual report counts an average workforce of 71,415 across industry, retail, finance and education. Its cooperatives elect their own management and keep pay differences narrow, by most accounts around six to one between the highest and lowest paid. It is not a utopia. A large share of the people who work in the group are employees and not worker-members: in its retail arm, about 8,300 of more than 28,000. Two of its larger cooperatives have voted to leave. But it has lasted seventy years and several recessions by letting the people who hold the information about the work also hold the decision and feel the result.
What these have in common is that they organise resources around what people actually value, because the loop from choice to consequence is short enough to feel. Command structures optimise for what can be transmitted up a hierarchy. Agency-led structures optimise for what the people inside them know.
Agency, Ownership, Feedback
The same logic works at the scale of one person, and here I will make the strongest claim in this essay. To get good work out of anybody, you need to do three things. Give them agency to act. Give them ownership of the outcome, good or bad. Give them a good feedback loop. That is all there is to managing people. The personality typologies, Theory X or Theory Y, Type A or Type B, do not change the recipe. Everyone I have worked with has responded to these three, and I suspect that is because they are what the small-group environment we evolved in provided by default.
Each one fails in a recognisable way when the other two are missing. Agency without ownership is chaos: people act and nothing comes back to them. Ownership without agency is blame: people are held responsible for outcomes they could not influence, which is a familiar experience at the bottom of large organisations. Agency and ownership without feedback is drift: people act, care about the result, and never learn whether they got it right.
Exhibit 3
Three ingredients, three ways to fail
| Given | Missing | What you get |
|---|---|---|
| Agency | Ownership | Chaos: people act and nothing comes back to them. |
| Ownership | Agency | Blame: responsibility for outcomes you cannot influence. |
| Agency and ownership | Feedback | Drift: effort and care, with no way to know if it worked. |
| All three | Nothing | Good work, from almost anyone. |
I have watched this hold in settings that have nothing in common. The team of six I ran at Camplus did their best work when each person owned a surface end to end and heard directly from the students using it. The student team I mentored to a win at the Joondalup Innovation Challenge, a local competition for solving civic problems, needed, above all, to be left alone to decide and then told honestly what landed. And the written working agreement I use with the AI agent that builds my solar modelling software, described in the previous essay, is the same three things written for a machine: the agent owns the route, it carries a duty to challenge and to report faithfully, and every edit, commit and turn closes a loop it cannot skip. Where one of the three was missing, the work degraded in exactly the way the framework predicts.
What I would and wouldn't claim
I would claim the definition of bureaucracy: loops that close on the inside. It explains more of what I have seen inside institutions than any account based on bad people or bad values. I would claim the three-part rule, and I would claim it strongly, because I have not yet found the person it does not apply to.
I would not claim that agency-led structures scale without cost, or that sortition should replace elections tomorrow. The Athenian pool was narrow, Mondragon's membership is partial, and citizens' assemblies work on one question at a time. What I would say is that every large organisation already contains small groups, and the choice is whether to design the loops those groups live inside or to let them emerge and hope.
Sources
- Hill, K. et al. (2011). Co-Residence Patterns in Hunter-Gatherer Societies Show Unique Human Social Structure. Science 331.
- Dunbar, R. (1992). Neocortex size as a constraint on group size in primates. Journal of Human Evolution 22(6).
- Lindenfors, P., Wartel, A. and Lind, J. (2021). 'Dunbar's number' deconstructed. Biology Letters.
- Bird, D. et al. (2019). Variability in the organization and size of hunter-gatherer groups: Foragers do not live in small-scale societies. Journal of Human Evolution 131.
- Gordon, D. (1999). Ants at Work. Summarised in her TED profile.
- Aristotle, Constitution of the Athenians 62.3, on serving twice on the Council. Kenyon translation. On the lot more generally: Britannica, Sortition.
- Guerrero, A. (2024). Lottocracy: Democracy Without Elections. Oxford University Press.
- The Citizens' Assembly (Ireland), overview. Referendum result: RTÉ, 26 May 2018.
- MONDRAGON Corporation (2026). Annual Report 2025. The pay ratio is reported by secondary sources and varies between six and nine to one; see Mondragon Corporation.
- McGregor, D. (1960). The Human Side of Enterprise, for Theory X and Theory Y. Friedman, M. and Rosenman, R. (1959), for Type A and Type B.