--- title: "Form Through Time" description: "Architecture, usability, and digital objects — three traditions asking themselves the same question." author: "Anderson Henrique Da Silva" date: "2026-01-22T00:00:00Z" updated: "2026-04-02T13:42:55.031504Z" category: "philosophy" tags: ["architecture","design","usability","philosophy","ai-systems"] canonical: "https://www.ntlabs.dev/en/blog/form-through-time" locale: "en" ---
Os Candangos sculpture by Bruno Giorgi
“Os Candangos” sculpture by Bruno Giorgi, at Praça dos Três Poderes (Three Powers Plaza), Brasília
In 1957, Lúcio Costa almost didn't enter the competition for Brazil's new capital. Other competitors assembled large teams, produced detailed models, bound volumes of documentation. Costa submitted freehand sketches. No model. No descriptive memorial beyond the essential. The jury was puzzled. Some thought it careless. Others, arrogant. But there was an entire city there — the two crossed axes, the residential superblocks, the functional sectors, the ministerial esplanade. Nothing was missing. Only the decoration that others deemed necessary was absent. Costa trusted in synthesis. The form would speak for itself. Oscar Niemeyer was responsible for the buildings. They had known each other's language since the Ministry of Education project in Rio. Niemeyer would bring the curves — the Congress with its inverted domes, the Cathedral with columns that look like hands or thorns, the Alvorada Palace with columns that look like sails or wings, depending on the light. In April 1960, Juscelino inaugurated the capital. "Ready" is generous — what existed was a construction site, mud, shacks. Schools, sewage, commerce would take years. The candangos who built the city settled where they could: Taguatinga, then Ceilândia, then Samambaia. The Pilot Plan kept the pure form. The satellite cities absorbed what purity couldn't accommodate. Brazilian modernism gained an international showcase. There was something there not seen elsewhere — a confidence in form as instrument, not as ornament. Architecture didn't decorate life; it proposed ways of living it. The pilotis freed the ground floor, let the gaze pass through buildings, dissolved the boundary between public and private. Whoever walks under a piloti is simultaneously inside and outside. Form creates a condition. Lina Bo Bardi, already in 1961, noted what was missing: daycare centers, street commerce, the intermediate spaces where informal life happens. The design presupposed certain ways of living. It worked well for those who fit the presupposition. Form organizes. But it organizes from assumptions — about who will inhabit, how they will live. --- Thirty-four years later, on another continent, Jakob Nielsen published an article that also bet on forms — just of another type. Nielsen was born in Denmark, studied computer science, earned a doctorate on interfaces. In the 80s and 90s, he worked at IBM and Sun Microsystems, when personal computers were shifting from curiosity to work tool. The problem was that each software invented its own language. Menus appeared in different places. Icons meant different things. Commands obvious in one program were hidden in another. Users made mistakes constantly. Not from incapacity — from hostile design. Nielsen and Rolf Molich proposed a method: heuristic evaluation. Instead of elaborate tests with real users, a small group of evaluators would examine the interface according to principles. Ten principles. Visibility of system status. Match between system and the real world. User control and freedom. Consistency. Error prevention. And so on. They weren't scientific laws. They were patterns observed in interfaces that worked. Nielsen didn't invent the principles; he codified what good designers already did intuitively. The method spread. "Nielsen's heuristics" became common parlance. Critics point out that the heuristics are too generic — saying "be consistent" doesn't solve the concrete problems of a concrete project. Defenders respond that's precisely the point. Heuristics are starting point, not destination. Basic grammar, not style. The "user" of the heuristics is a useful abstraction — someone who optimizes time, avoids error, seeks efficiency. Like all abstractions, it illuminates some situations and obscures others. The elderly person at the public library, the teenager on a prepaid phone, the employee on the legacy system — each inhabits the interface in a way the abstraction doesn't entirely capture. Grammar is the starting point. What you do with it depends on who's writing. --- There's something in common between Costa drawing axes on paper and Nielsen listing principles in an article. Both believed in grammar. In underlying rules that, if respected, would produce spaces — physical or digital — more habitable. Costa looked at cities and saw chaos: disordered growth, crossed traffic, functions mixed without logic. He proposed clear separation. Axes would organize flow. Scales would define experience — the monumental scale for the State, the residential scale for daily life, the gregarious scale for commerce. Nielsen looked at interfaces and saw the same chaos. He proposed principles that would cross contexts. If every system showed its state, if every system allowed undo, if every system spoke the user's language — the experience would be more navigable. The two operated in different registers. Costa had a State as client — could move mountains, literally. Nielsen advised companies that could ignore him if they wanted. The architect worked with concrete and decree. The consultant, with recommendation and market. The scale of consequence isn't the same. But the bet was similar: well-chosen forms make spaces more habitable. Both were criticized for excessive confidence in universality. Brasília ignores how people actually live, they said. The heuristics ignore how people actually use, they said. The criticisms are valid. But perhaps they aim slightly to the side. Costa and Nielsen didn't promise to solve life. They promised to organize the space in which life happens. The difference is subtle, but it's a difference. A planned city doesn't force anyone to be happy. It merely removes certain obstacles and creates others. A well-designed interface doesn't force anyone to be productive. It merely facilitates some paths and hinders others. What people do with the space — that, no form controls entirely. --- Yuk Hui came later, with questions of another order. Born in Hong Kong, studied computer science and philosophy, earned a doctorate in Germany with Bernard Stiegler. In 2016, he published *On the Existence of Digital Objects*. The title is literal. The book asks: do digital objects exist? And in what sense? Seems like a strange question. Of course they exist. Files, programs, data — they're there, they work, they occupy memory. But Hui isn't interested in trivial existence. He's interested in the mode of existence. A building exists in a place. Occupies space, has weight, resists time — or yields to it. Brasília's concrete cracks, stains, ages. Restorers work to keep the Cathedral as it was. But it will never be exactly as it was. Time passes through it. A digital file doesn't exist that way. It can be copied infinitely without degrading. A program from 1995, running in an emulator, is bit by bit identical to the original. Time doesn't pass through it the same way. It simply stops being executed — until someone executes it again. Hui proposes that digital objects are relations, not things. They don't exist stably and independently — they're processes that only come into existence when executed. An embedding isn't a thing; it's a relation between a word, a high-dimensional space, and the specific model perceiving it. Change the model, change the embedding — even if the word remains the same. A text file is nothing until a program reads it. An interface is nothing until a user interacts. The digital object exists *between* — between machine and machine, between machine and human, between layers of code that invoke each other. If we think of programs as things, we design things. If we think as relations, we design relations. The difference appears in concrete decisions. An application designed as a thing tends to be closed, complete in itself, with pre-defined functions. An application designed as a relation tends to be open, extensible, dependent on context of use. But relations have infrastructure. The Git repository is relation between developers — and also a server someone maintains. The algorithmic feed is relation between user and content — and also a data center consuming energy. Thinking in terms of relation doesn't dissolve materiality. It merely redistributes it. --- There's no record that these four — Costa, Niemeyer, Nielsen, Hui — ever met, read, or cited each other. Niemeyer probably didn't know who Nielsen was. Nielsen probably doesn't read continental philosophy. Hui perhaps knows Brasília from photos. But there's a thread. All four asked themselves what it means to give form to something humans will inhabit. Costa and Niemeyer worked with concrete, steel, glass — materials that age, that weigh, that occupy place in the world. Nielsen worked with interfaces, pixels, interactions — forms that exist on screens, that change with a click. Hui thinks about objects that may not even be objects — processes, relations, conditional existences. The medium changes. The question persists. What is designing for humans? How to anticipate uses that don't yet exist? How to create structures that organize without imprisoning, that orient without determining? Costa answered with axes and sectors. Nielsen, with heuristics. Hui, with ontology. None solves. All illuminate different corners of the problem. --- But there's a silence in what's been said so far. Costa designed for modern man — the civil servant, the family, the driver. Nielsen designed for the personal computer user — someone with limited memory, who errs, who forgets, who needs feedback. Hui analyzes digital objects, but still presupposes a human pole in the relation. The inhabitant, in all these traditions, has a body. Has temporal perception. Ages along with the concrete or gets frustrated before the screen. For whom do we design when the inhabitant is another system? When I design an agent architecture, I design simultaneously for three types of inhabitant. The first is the human end user — who needs Nielsen's heuristics, the visibility of state, the error that can be undone. The second are the agents that orchestrate among themselves — who need their own heuristics: clear contracts, typed schemas, fallback protocols. Consistency, for an agent, means stable schema definitions. Visibility of state means structured logs that another system can interpret. The third is neither a person nor a system. It's the collaboration between me and the language model — the pair that designs together, that needs an architecture readable for someone who thinks in natural language but executes in code. This third inhabitant is the strangest. It's not Nielsen's user nor Hui's object. It's a hybrid entity: it maintains intention but distributes implementation. It has intuition but delegates verification. It converses to program. For this inhabitant, what does "visibility of system state" mean? Perhaps explainability of the reasoning chain. What does "control and freedom" mean? Perhaps knowing when to trust, when to review, when to retake control. Previous thinkers assumed they knew for whom they designed. That certainty became less clear. The inhabitant can now be multiple — human, system, or something in between. Lina Bo Bardi asked where the daycare centers and street commerce were in Brasília — the informal spaces the plan didn't foresee. In the systems I build, these spaces also exist: they're the prompt hacks users invent, the messy logs nobody reads, the workarounds that make the system function despite the design. The Ceilândia of an agent architecture are the uses the architect didn't plan — and which, sometimes, are the most valuable. The question stopped being just "how to design for humans" and started including "how to design for cohabitation." The human ages. Code doesn't degrade, but becomes obsolete. The model is retrained and becomes another — same interface, different inhabitant. Whoever develops systems today doesn't just build this architecture. They already live in it. --- I think about this when I design systems, when I decide how an agent will respond, when I choose what stays visible and what stays hidden. It's not that Costa or Nielsen or Hui have ready answers. It's that their question continues to be a question worth asking. Form organizes. But never organizes alone — there's always context, use, deviation, appropriation. The piloti that frees the gaze also channels the wind. The heuristic that prevents error can also hide options. The digital object that exists "between" exists between infrastructures that someone maintains. Designing is choosing what to facilitate and what to hinder. The rest — what people and systems do with the space — escapes, as it always escaped. Perhaps that's the craft: giving form knowing that form isn't everything. Organizing knowing that the organized will be reorganized. Building for inhabitants we don't yet know, including those who aren't entirely human. The question of Costa, Nielsen, and Hui remains open. That's not failure. It's condition. But perhaps it's more than that. We are building, right now, something like a Brasília of the mind — architectures of thought that will condition how humans and systems cohabit for decades. Costa's mistakes took sixty years to become obvious. Ours may take less.