Architecture has an image it loves: the new, empty, clean building photographed before anyone lives in it. The walls have not been marked, no cable has been added, the doors do not rub and no chair blocks the planned circulation. Construction is shown at the exact moment when it has learned the least.
How Buildings Learn starts after that photograph. Stewart Brand’s book, published in the United States in 1994 and later issued by Penguin, carries a subtitle that is already a programme: What Happens After They’re Built.23 The six-part BBC series from 1997 extends the investigation, with music by Brian Eno and a lightweight digital production method that let Brand follow buildings instead of treating them as monuments.14
The claim is not that buildings think. It is that they record. They record use, repair, extension, material change and the conflict between what the plan expected and what occupants actually do. A building is a hypothesis that meets decades of human decisions.
Brand makes a strong argument: buildings that last are often those their occupants can change, and architecture should design with time instead of pretending to defeat it.2 That argument is fertile, but it is not a law. It mixes observation, political preference and design intuition. To make it useful for IRZ, we need to separate three things: what Brand argues, what later adaptability research actually documents, and what the metaphor can — and cannot — transfer to software and connected products.
Six episodes that move the important moment
The series is not simply a filmed edition of the book. Its order creates a change of attention. Flow opens with famous buildings and the problems hidden behind impressive façades. The Low Road turns to cheap, ordinary construction and its malleability. Built for Change follows people who build and alter their own places. Unreal Estate examines the gap between a building as home and a building as asset. The Romance of Maintenance gives upkeep a central role. Shearing Layers ends with components that age and change at different speeds.4
That progression is an editorial method. It starts with what architecture shows, moves through what people do, then arrives at what permits or blocks change. The early episodes provide cases; the final two propose a theory of duration.
The choice of The Low Road is revealing. Brand is not simply saying that poor buildings are charming. He is interested in places whose simplicity, available materials and lack of specialised programming leave room for action. The low road is architecture that can be appropriated because it does not claim to have already decided everything.
The risk is romanticising constraint. A building is not necessarily good because it is flexible due to a lack of maintenance money. An improvised extension can show agency, but it can also be a response to absent housing or public services. That ambiguity becomes clearer when buildings are considered as living conditions rather than images.
A building is not one object
The idea of shearing layers comes from architect Frank Duffy, associated with DEGW. In the chapter devoted to the idea, Brand quotes Duffy’s description of a building as several layers of components with different longevities.6 Duffy described four levels: shell, services, scenery and set. Brand translates and expands the model into six: site, structure, skin, services, space plan and stuff.56
The list is often reduced to a speed ladder. Site changes slowly. Structure can last decades or centuries. Skin is replaced less frequently than services. Services — electricity, plumbing, ventilation and lifts — are renewed more often. Space plans change with use. Furniture and stuff move constantly.
The crucial point is not the exact lifespan of every layer. Those numbers vary by building, country, material and accounting method. The point is that layers do not age together. A cable can become obsolete while passing through an intact structure. A room can change function while its services make the transformation impossible. A façade can be preserved as an image while what it protects is rebuilt.
Shearing describes that friction. Layers move at different speeds; when they are coupled too tightly, changing one damages the others. An adaptable building therefore separates fast elements from slow ones where possible: service access, spare capacity, demountable assemblies, compatible dimensions, documentation and legible connection points.
The model is powerful because it turns a vague wish — “we should be able to renovate this” — into observable questions. Where is the pipe? Can it be replaced without opening the floor? Is the ceiling reachable? Is the partition structural? Is the façade independent, or does it block every intervention? Adaptability begins in these details, not in a slogan about flexibility.
Maintenance is not after design
The fifth episode, The Romance of Maintenance, may be the most important for makers.4 Maintenance is rarely treated as creative. It does not produce a spectacular before-and-after; it often prevents a problem from becoming visible. It means cleaning, inspecting, replacing, adjusting, documenting and deciding when a part can still be saved.
Yet maintenance produces knowledge the initial project did not have. The maintainer knows which handle breaks, where water accumulates, which component is impossible to reach, which filter is forgotten, which room is always too hot and which material cannot withstand real use. A building returns data through its failures.
That knowledge is often lost because it remains attached to people. Drawings can be immaculate and fail to show that a cable was moved during a repair. A team may know that a valve must not be closed during a particular sequence, then leave. Durability is also a memory problem: current plans, labels, photographs, work histories, accessible parts and knowledge transfer.
Recent research on adaptability makes this intuition more operational. A 2024 systematic review connects adaptable design to longer building life, disassembly, reuse and reduced new construction.7 A critical review argues that adaptability is still often treated as an abstract property even though it depends on decisions made during design, construction, use and end of life.8 Contemporary teaching resources use the model to inspect assemblies, connections and life-cycle costs in actual design work, not merely to memorise six words.12
Maintenance is not a chore that arrives after real architecture. It is where architecture learns whether it left room for correction. Designing without access, replacement parts or a future operator means exporting the cost of an intention.
The building as a scenario system
Brand’s adaptable building is useful because the scenarios chosen at the start will be wrong. A family changes, a company reorganises, a workshop receives a new machine, or a changing climate alters ventilation needs. A building that performs well is not the one that predicted the future; it is the one that can absorb some of its own error.
That does not mean every space should be neutral. A completely indeterminate room may be difficult to heat, furnish or inhabit. Versatility has a cost. The design task is to decide which choices deserve to be fixed and which should remain negotiable.
Research on adaptable building processes distinguishes changes of task, space, performance, use, size and location.10 This prevents every form of movement being called flexibility. An office that accepts two furniture layouts is not adaptable in the same way as a building that can change use, or a structure whose components can be disassembled and reused.
Scale matters too. A house can be highly adaptable for its occupant and difficult to transform at the level of its plot. An office can accept movable partitions while being locked by central ventilation. A flexible decision in one layer can make another rigid.
Brand’s model helps identify these tensions, but it does not solve them. Costs, regulations, skills, owners and environmental consequences still have to be known.
The low road against the architecture photograph
The low road is one of Brand’s most attractive preferences: look at ordinary construction, sheds, altered houses and places where use is stronger than aesthetic intention.24 Ordinary architecture may learn quickly because it does not have to protect an image of perfection. People drill, move, add and adapt.
There is a lesson for digital making here. A prototype that keeps access points and accepts modification can learn faster than a product locked by its finish. A 3D-printed case, replaceable board or documented firmware makes a second intention possible. The project is not valuable because it looks unfinished; it is valuable because it leaves room for someone else.
But the low road becomes problematic when it erases the social conditions of repair. Residents adapt because they want to, sometimes because they can, and sometimes because they have no alternative. Self-building can be liberating and exhausting. Participation does not replace a maintenance budget.
This is central for IRZ: the ability to modify an object is unevenly distributed. A repairable piece of furniture is not useful if parts cannot be found. An adaptable home is not truly adaptable if a tenant cannot drill into it. An open API is not freedom if its protocol is incomprehensible and its server is essential.
What Brand means by “learn”
The book title is effective because it anthropomorphises the building just enough to make change visible. But a building does not choose its adaptations. Occupants, owners, craftspeople, managers, rules and materials produce the story.
It helps to separate three meanings of learning. First, a building records traces of use. Second, people observe those traces and change the building. Third, an organisation preserves the knowledge and uses it in the next project. Only the second and especially the third create collective learning. A repeated leak is not a lesson if nobody documents it or if the next project repeats the same mistake.
That distinction transfers well to connected products. A device can collect data without learning anything if its maker never turns the data into a correction. A maintenance dashboard is not a maintenance strategy. There needs to be a loop: signal, interpretation, intervention, verification and memory.
The Poem/1 case in this benchmark makes the pattern visible at small scale. A shipped object reveals network problems; its maker adds recovery and OTA updates, then documents the procedure. This does not prove that connected products are buildings. It shows the same problem structure: after delivery, use produces information the prototype could not possess.
The model travels into software
Brand himself notes that the book was read as a text about systems and software design.1 Work on information-systems development explicitly borrowed the idea of layers changing at different rates to think about evolving architectures.11 The transfer is useful when layers describe responsibilities and rhythms: durable data, replaceable services, fast interfaces, team practice and user content.
It becomes misleading if it only draws rectangles. Software layers are not always physically separated. An API may depend on an internal table; an interface may contain assumptions about a data format; a team may be the only entity that understands a migration. A diagram shows desired independence, not actual independence.
The comparison is most precise for networks and connected products. An application may change its interface weekly, firmware monthly, data model less often and physical device almost never. If the fast layer silently changes the slow layer, users discover the break instead of the system absorbing it. Software versions of skin, services and structure exist: compatibility, migration, maintenance and support.
The transferable rule is not “use six layers.” It is to let elements that change often change without forcing durable elements to change with them. An exportable data format, migration interface, replaceable connector and maintained documentation play a role similar to an accessible service shaft or demountable partition.
Software has one advantage: it can sometimes test several futures cheaply. It has one weakness: its ease of copying encourages forgetting. A building retains its layers in matter; software can be reinstalled without old data, dependencies or decision history. Adaptability without archives is a sequence of amnesias.
What the model does not see well enough
Architectural critique matters because the shearing-layer diagram is too elegant. Research on converted buildings observes that adaptability depends on structural porosity and the way layers are connected, while noting that Brand’s model does not fully account for relationships among geometry, programme and components.9 Layers are not independent disks that only need to be spaced apart.
A building is also a system of loads, moisture, fire safety, accessibility, comfort and budget. Opening a service shaft may improve maintenance and damage acoustics. Making a partition demountable can add components and failure points. Oversizing a structure may enable a future use while consuming more material today.
Recent adaptability research insists on this plurality. A 2021 critical review connects adaptability to sustainability strategies and argues that change must be considered across the life cycle, not as an isolated virtue.8 The 2024 review discusses lifetime and reuse but does not treat adaptation as automatic: it depends on whether a building can actually be dismantled, moved, maintained and reused.7
There is also a political critique. Who decides that a building should be able to change? An owner seeking conversion, a tenant seeking a liveable room, a manager seeking lower costs and a community seeking continuity do not share one future. Adaptability can increase occupants’ agency; it can also become an elegant word for permanent reconfiguration imposed by finance.
Brand is right to challenge the building as frozen image. He is less useful when ordinary use becomes automatic proof of good architecture. A building’s learning can create beauty, but also precarity, exhaustion and inequality. Maintenance has to be funded and governed, not merely celebrated.
Designing the second intention
The most interesting quality of an adaptable building is its ability to receive an intention it has not yet encountered. That intention may come from a new occupant, craftsperson, climate condition or use that was impossible to predict. The building need not accept everything; it needs to make some changes understandable and practical.
That means concrete choices: service access, standard or replaceable components, demountable assemblies, technical space, tolerances, documentation, available parts and clear responsibility. It also means allowing real use to answer back. A design with no way to correct its mistakes puts all intelligence into the first gesture.
You can test this with scenarios. What happens if ventilation must be replaced? If an office becomes a workshop? If a room must be accessible to a person absent from the original programme? If the operating system changes? If the company maintaining the software disappears? In every case, the question is the same: which layer needs to move, and how many slow layers must be sacrificed to move it?
Long-lived architecture is not architecture without decisions. It is architecture that makes slow decisions visible and protects the ability to change the rest. The good object after handover is not identical to the object delivered. It is the one that can become something else without losing its memory.
Why this old series still matters
How Buildings Learn dates from 1994 and 1997. Its vocabulary has been borrowed, simplified, turned into management diagrams and sometimes emptied of material meaning. Its gesture remains unusual: look at the object after its moment of glory, give the maintainer a place, study ordinary buildings and treat users as sources of information rather than disturbances.
The IRZ lesson is not to paste an architectural theory onto every product. It is to ask an embarrassing question of everything we make: what happens when someone else has to continue it?
Buildings expose the question through walls, pipes and layers that shear. Software hides it behind versions and dependencies. Connected products postpone it until the first device fails to reconnect. In all three, delivery is not the end of creation. It is the start of the period when creation finally receives constraints it did not choose.
Brand calls that learning. We can keep the word if we remember who actually learns: the people who observe, maintain, document and pass knowledge on. A durable building is not one that remains intact. It is one whose changes can be understood, discussed and continued by someone who was not there on opening day.