Every design settles ahead of time who will have to adapt when things go differently than planned. Anything designed embodies expectations about reality, whether it is an interface, a building, a bureaucracy, a workflow, a law, a workplace, a classification system or a transit network. Because reality will never perfectly match those expectations, design determines what happens when the mismatch appears, including what stays fixed and who has to bend around it.
That makes the design question political before anyone argues about politics. The visible instruments of power have not gone away: statutes still pass, courts still rule, and the police still exist. But the layer that decides whether an ordinary week works has moved beneath them, into the friction of administrative interfaces, digital workflows and algorithmic gatekeeping. An agency or employer rarely has to issue an order anymore. It is enough to route the appointment through a portal, place the appeal in a queue, and let the person on the other end do the absorbing.
Design relocates friction
Simplifying one actor’s experience often means somebody else absorbs complexity. The important design question is less “have we eliminated friction?” than where it went. A self-service system that reduces the institution’s clerical cost converts customers into unpaid clerks. A lean schedule that maximizes throughput converts demand uncertainty into worker precarity. The interface that feels effortless may be effortless precisely because the burden was moved.
Standardization creates a remainder
Schemas privilege some cases as normal and turn others into exceptions. How a design treats those exceptions is where much of its ethics shows. Must the person contort themselves around the system, can the system accommodate them, or can repeated exceptions alter the underlying standard? The remainder is information about what the design presupposed, and it belongs inside design rather than after it.
Good design includes its own correction
The fantasy of anticipating everything is itself dangerous, because it encourages designers to treat unforeseen cases as user error, which files the design’s own defect under the name of whoever ran into it. Since mismatch is inevitable, mature design must include mechanisms for its own correction. Good design is competent initial design plus the capacity for redesign. With that, corrigibility, whether a system can be corrected by the people it gets wrong, becomes a design question as well as a governance one.
Design redistributes discretion
A rigid system often moves discretion instead of eliminating it. Remove frontline discretion and users may devise workarounds. Automate judgment and somebody must handle exceptions. Standardize a workflow and informal repair proliferates around it. So the question goes beyond “how much discretion?” to who receives the burden and authority of dealing with reality when the specification fails.
Design keeps deciding after the designer leaves
A designer does not have to command every future user. The artifact keeps enforcing decisions after the designer is gone. A building entered only by stairs keeps excluding wheelchair users long after its architect has moved on, and each of them has to find another way in. Choices like that come to look like facts about the building, which later users and managers learn to work around. That is one reason design can depoliticize power — “that’s just how the system works.” The framework’s response is that someone decided which part of reality the system would treat as fixed.
Closed design and corrigible design
Closed design says, implicitly, that reality must conform to the specification. Corrigible design treats the specification as a provisional hypothesis about reality, which is a quite different design philosophy. A hypothesis needs someone entitled to test it. The people the mismatch lands on should have standing to put the specification itself under review, beyond their own case, or the capacity for redesign stays in the designer’s hands alone.
Politics moved into the workflow
The same relocation is now the central political conflict, and it is already running on five fronts. Each has the same shape: an institution records smooth operation while the people inside it pay in unrecorded capacity, and the gap between the two is where the fight has started.
The liability sponge. Generative models are being pushed into consequential work on a corporate timetable that has little to do with their reliability: prior authorization, clinical documentation, benefits processing, hiring screens, legal discovery. As the model’s errors surface, the fixes that would make it trustworthy turn out to be slow and expensive, so the fast fix is a human — an underpaid reviewer placed in the loop to initial hundreds of machine drafts. The reviewer is there so that, when the denial is wrong or the order is unsafe, the complaint has a licensed person to name, the model being hard to depose. That turns the dispute over whether AI safety is an ethics program or an engineering requirement into a labor fight. Whoever holds the signature holds the liability, so whoever signs has a reason to strike. The Unpaid Fixes Are Already in Your Logs works through how the signature moves that liability off the institution’s books.
The dashboard and the day. Employment, productivity, digital take-up: the macro numbers look good. The micro experience is an unpaid second shift of administrative work — hold music, password resets, re-uploaded documents, claim forms that ask for facts the agency already has — spent proving eligibility for things that were supposed to be entitlements. The Metric She Repairs Says She Is Not Needed follows one scheduler whose entire afternoon goes into repairing a dashboard’s numbers, and The Official Record Is Late explains why the record cannot see her. When enough people live inside that gap, the politics changes register. The anger skips corruption and targets insulation: nobody in the chain exercised arbitrary will, and nobody can be corrected, so the grievance lands nowhere and the person holding it keeps it. Grievances like that do not stay private. They feed anti-system politics of every flavor, and the honest description is not conspiratorial: the systems genuinely run, and they genuinely run on people.
The strike over unbilled repair. The newest labor disputes are about something other than pay. Nurses are striking over ratios that assume a charting system made them faster, when the system gave them duplicate screens and the evenings to feed them. Teachers are burning out managing tracking apps and compliance portals off the clock. The demands forming around these jobs aim at containing the work the software creates: staffing floors, recognition of the off-hours charting the industry calls pajama time as the compensable labor it already is, and the right to refuse an unvalidated rollout. The Capture of “Maintenance” names the asset being fought over; the bargaining language, including the off-hours figure, lives in The Cost of Being Heard.
Attrition as budget policy. In the welfare state, automation arrives as a way to say no at scale. A fraud detector that flags the wrong families, a review clock that resets whenever a form moves, a queue with no deadline: each cuts spending without ever issuing a denial, because the saving is collected from the people who stop calling back. Robodebt ran on that arithmetic and Stop the Machine, Not the Person audits it; Pending Is a No Nobody Signed names the queue itself. What is forming against it is a civil-rights claim in administrative-law clothing: the right to a statutory deadline after which silence turns into approval. A deadline like that converts waiting from a weapon into a line item. Administrative latency has been a budget line. Making it one that bills the institution is the fight.
Care theater, on both sides. One side answers broken services with the language of care — statements, task forces, empathetic chatbots — and Sanctiphagy (Virtue Extraction) describes that swap: praise for workers in place of hiring more of them. The other side answers the same broken services by attacking the language, promising deregulation, which swaps public friction for private extraction and changes nobody’s hold music. Both leave the operational layer untouched, which is why alternating the party in power changes the vocabulary and rarely the week. The extraction lives downstream of protocols rather than personalities, and for the specific mechanisms named here — portals, queues, sign-off screens — the suspicion that elections have not touched them is accurate. Statutes, courts and police still matter, and procurement and administrative-law disputes are where the fight is already running.
Underneath the five disputes sit three questions: who pays the cost of the friction, who holds the authorization to change it, and who decides whose time and health get spent to keep the system running. Which dispute to take up first is a separate question, and the intuitive answers are mostly wrong. Whichever comes first, it will be settled in material terms — strike demands, procurement clauses, administrative-law dockets — and those are already being contested.
Who adapts first
The ordering is short. An institution should spend its own capacity before anyone else’s: absorb complexity in its own systems first, and ask people to adapt only for what no design can remove. The Capacity Agenda works through what that ordering asks of policy.
Good design spends the institution’s collective capacity so that people get their own back. When the institution absorbs complexity, the people it serves spend fewer evenings on hold and have more of those evenings left for living.
A well-designed institution fits itself to people’s circumstances wherever it can, so that they spend less of their lives fitting themselves to it. Who adjusts to whom then becomes something designers decide on purpose, and something the people they serve can ask them to justify.