Article
Ulrich’s boundary critique, via Bradbury’s automated house: why a system can meet every target it was set and still be making breakfast for nobody.
By Richard Faint · 4 August 2026 · 11 min read
Bradbury’s automated house meets every target it was designed for while its human purpose has disappeared. The lesson for product managers is to question system boundaries and intended beneficiaries, not just whether the system is operating efficiently.
“Today is August 4, 2026.”
On that date, a house in Allendale, California, wakes its occupants. It announces the time, prepares breakfast, cleans the rooms and reminds the family about birthdays, bills and appointments. Robotic mice emerge from the walls to remove dust. The house controls the temperature, waters the garden, prepares meals and selects an evening poem. It is coordinated, responsive and almost completely autonomous. There is only one problem. As Bradbury tells us near the beginning of the story, “The morning house lay empty.” The family is dead, the surrounding city has been destroyed by nuclear war, and the machinery is continuing with a routine whose purpose has disappeared.
Ray Bradbury’s short story There Will Come Soft Rains was first published in 1950, the date on which it takes place has arrived. Bradbury imagined something remarkably close to the modern smart home: voice-controlled systems, automated lighting, climate control, robotic cleaning, scheduled notifications, domestic sensors and machines capable of preparing food. None of those technologies now seems especially extraordinary. What makes the story unsettling is not that Bradbury predicted them, tt is that they continue to work perfectly when everyone is dead.
The house continues to execute its routines because that is what it was designed to do. At seven o’clock, it tells the family to wake. At breakfast time, the kitchen produces eggs, bacon, toast, coffee and milk. Later, it clears away dirt, waters the garden and prepares the rooms for the evening. Every subsystem performs its assigned function. The clock keeps time, the stove cooks, the cleaning machines remove waste, and the house coordinates its components without direct human intervention.
It cannot understand that the family is gone, it cannot recognise that the context giving those activities meaning has disappeared. It cannot ask whether breakfast still needs to be cooked, whether the children still need reminding about school, or whether the evening cigar should still be placed beside an empty chair. It has retained the process while losing the purpose.
This is what makes the house more than an early prediction of Alexa, Roomba or the Internet of Things. It is a system optimising its internal processes while remaining blind to the collapse of the wider system around it. The house is efficient, reliable and technologically sophisticated, it is also useless.
Organisations regularly make the same mistake on a less catastrophic scale. A process can run smoothly while producing no meaningful value. A department can achieve every target while damaging the wider operation. Bradbury’s house is an extreme version of a familiar organisational failure: a system continuing to do exactly what it has been instructed to do long after those instructions have stopped making sense.
It is tempting to treat the story primarily as evidence that Bradbury predicted modern domestic technology. He certainly anticipated much of it.
The smart home, however, is not the real warning. The city has been destroyed by nuclear attack and the family survives only as silhouettes burned onto an outside wall. A man is shown mowing the lawn. A woman reaches for flowers. A boy throws a ball towards a girl whose hands remain raised to catch it. The technical system can manage the smallest details of domestic life, but the political and military systems surrounding it have failed completely.
Technology has advanced. Human judgement has not.
Bradbury was rarely interested in machinery simply as machinery. His work repeatedly examines what technology reveals about the people who build and use it. The automated house is impressive, but it cannot protect its occupants from decisions made beyond its walls. It can regulate the temperature of a nursery but cannot regulate the international system capable of destroying the child who slept there.
A boundary determines what we treat as part of the system and what we classify as its environment. It influences which relationships are examined, which outcomes are measured, which people are consulted and which consequences are treated as external. The boundary is not necessarily a physical line found in the world. It is an analytical choice made by the observer, and that choice shapes the conclusions that follow.
Draw the boundary tightly around Bradbury’s house and it appears successful. Its internal components remain coordinated, its routines are completed and its resources are deployed according to plan. Draw the boundary around the household and the assessment changes because the services are being performed for people who no longer exist. Draw it around Allendale and the house becomes a functioning fragment inside a destroyed city. Draw it around human civilisation and the absurdity becomes complete: humanity has mastered domestic automation while failing to prevent its own extinction.
Nothing inside the house changes as the analytical boundary expands. What changes is our judgement about whether the system is successful. Churchman argued we are “always obliged to think about the larger system.” This does not mean that every analysis can include everything. Any useful model must exclude most of reality. It means that those exclusions should remain visible and open to challenge. A solution that appears effective within one boundary may transfer costs, risks or failures into another part of the wider system.
In logistics, a warehouse may improve its internal picking efficiency by releasing work in larger batches, while creating queues in the yard and missed delivery slots for transport. A route optimiser may reduce mileage while creating plans that drivers cannot complete safely. A product team may increase feature throughput while support demand, technical debt and user confusion rise elsewhere. The local measure improves because the cost has crossed the boundary of the team or model used to assess it.
Bradbury’s house is perhaps the ultimate local optimum: a perfectly functioning home in a world where nobody remains to live in it.
Ulrich developed this problem through Critical Systems Heuristics, an approach centred on boundary critique. Ulrich’s central argument is that claims about improvement always depend on prior judgements about the relevant system. Before something can be described as better, successful or efficient, somebody must decide what counts, who counts and which effects fall outside the assessment.
Boundary critique asks questions that conventional analysis often leaves implicit. What is the system supposed to achieve? Who is intended to benefit? Who bears its costs and risks? Whose knowledge is treated as authoritative? Who has the power to define success? Who represents people who are affected by the system but not involved in designing or governing it?
These are not secondary ethical questions to be added after a technical solution has been designed. They are part of the design itself. Every objective function, requirements document, dashboard and operating model embeds assumptions about what matters. Every model excludes something. Every optimisation method defines some outcomes as valuable, others as constraints, and many more as irrelevant noise.
The problem is not that boundaries exist. Analysis would be impossible without them. The problem appears when a chosen boundary is presented as though it were natural, complete or neutral. A system may appear highly successful because dissatisfied users, exhausted employees, environmental damage or long-term maintenance costs have been placed outside the model. Change the boundary and the same performance can look very different.
Bradbury’s house has an exceptionally narrow operational boundary. It understands rooms, schedules, temperatures, movement and household routines. Within that boundary, its behaviour remains rational. The absence of the family is initially little more than a missing response. Nuclear destruction is outside the model.
We can also distinguishes between systems that can be treated as relatively closed and those that depend on continuing exchanges with their environment. Organisations and technological platforms are open systems. They rely on flows of energy, information, materials, maintenance, money, labour and institutional support. They may appear autonomous, but their operation depends on networks extending far beyond their visible boundary.
Bradbury’s house depends on electricity, water, physical infrastructure and human maintenance. For most of the story, those dependencies remain invisible because some supporting resources are still available. The house appears self-sufficient only because the wider systems sustaining it have not yet failed completely.
The same illusion surrounds many modern digital services. A customer interacts with a clean interface that appears instantaneous and automatic, but that experience depends on data centres, electrical grids, telecommunications networks, software libraries, cybersecurity teams, supply chains, engineers, regulators and outsourced labour. Artificial intelligence appears to produce an answer by itself, although the service rests on extensive technical, economic and human infrastructure.
Automation does not remove dependencies. It conceals them.
The more seamless the service becomes, the easier it is to forget the systems and people required to sustain it. This produces a particular form of organisational fragility. When the hidden infrastructure fails, the supposedly autonomous service can collapse rapidly because users and managers have lost sight of how it actually works.
Optimisation requires an objective function: something the system is instructed to increase, reduce or maintain. That objective is only a representation of the real purpose. It is never the purpose itself. Distance may be used as a proxy for transport efficiency, closure time as a proxy for customer service, output as a proxy for productivity, and engagement as a proxy for value. Each can be useful, but each captures only part of the system.
Once the proxy becomes the target, behaviour reorganises around the measure. The distinction between the metric and the underlying purpose gradually disappears. Teams learn how to improve the number, algorithms become better at optimising it, and governance structures reward the reported outcome. The wider purpose is assumed rather than examined.
The house’s routines once represented care for the family. Breakfast, heating, cleaning and reminders all contributed to a functioning home. After the family’s death, those same routines continue, but their relationship to the original purpose has been severed. The activities remain. The value has disappeared.
This is a particularly important warning for AI adoption. Automating an existing process can make it faster without establishing whether it is useful. Adding intelligence to a workflow can reinforce its assumptions rather than challenge them. A poor process executed by people may at least generate friction, complaints and visible signs of failure. A poor process executed automatically can operate at enormous scale while appearing technically successful.
Before asking whether a process can be automated, organisations should ask why it exists, who benefits from it, which wider outcomes it supports and what evidence would show that it is no longer useful. Otherwise, they risk building increasingly sophisticated houses that prepare breakfast for nobody.
In the evening, the house asks its absent owner which poem she would like to hear. There is no response, so the system selects Sara Teasdale’s There Will Come Soft Rains. It imagines nature continuing after humanity has destroyed itself. Birds, frogs, trees and spring remain indifferent to human extinction.
The house can recite the poem but cannot understand its relevance. It possesses the text, pronounces the words and follows the instruction to provide entertainment, yet it cannot connect the poem’s meaning to the empty rooms, the ruined city or the silhouettes on its wall. It can retrieve language without recognising the warning contained within it.
Read in 2026, the scene inevitably brings generative AI to mind. A system may produce persuasive language about grief, war, responsibility or extinction without experiencing any of them. It can identify relationships and construct a meaningful response without possessing human stakes in the outcome. That does not make the output worthless, but it means that interpretation, judgement and responsibility remain with people.
The house can read the warning. Only people can choose to act on it.
Eventually, the house catches fire and it responds immediately. Alarms sound, water pours from the ceilings, doors close and automated devices rush to defend the building. For a time, the house appears almost alive, fighting desperately to preserve itself. Its response is coordinated and technically impressive, but its resources are finite. The water supply fails, the fire spreads and one automated function after another collapses until the structure is destroyed.
The system that survived its creators cannot survive indefinitely without them. Its apparent independence was temporary, supported by infrastructure it did not control and could not repair. Near the end, even the house recognises that “only silence was here,” but recognition arrives too late and changes nothing.
The ending adds another layer to Bradbury’s warning. Technological systems may create the appearance of autonomy while remaining dependent on larger physical and social systems. A smart system is never truly self-contained. Its autonomy rests on networks that often remain invisible until they fail.
The most striking thing about reaching Bradbury’s date is not how much he predicted correctly. It is how familiar his central problem remains. We are surrounded by systems designed to make activity faster, cheaper and more automatic. We can automate communication, planning, analysis, recruitment, logistics and decision support. We can build systems that continue operating without constant human involvement.
Automation does not remove the need to ask what the system is for. Nor does technical sophistication remove the need to examine where its boundary has been drawn. A system may appear successful because its users, employees, environmental costs or long-term consequences have been excluded from the model. A department may appear efficient because the work it creates for other departments is not counted. An algorithm may appear accurate because the people harmed by its errors are absent from the primary measure.
Change the boundary and success can become failure.
Efficiency is not purpose. Activity is not value. Intelligence is not wisdom. A system can perform every task correctly and still contribute to the wrong outcome.
On August 4, 2026, Bradbury’s automated house wakes up, prepares breakfast and announces the beginning of another day. Nobody answers. The machinery continues regardless.
That is the part of the prediction we should pay attention to.
Bradbury, Ray. There Will Come Soft Rains. First published in 1950 and later collected in The Martian Chronicles. Read the story.